A Tribute to Bob Pease Pease Porridge —1 (1990-1992) Wonderful World of W/F's. Wika peepee Werner he doce snanr Ye marriage! inguion Waid a Pyeeng Uesepe aici Stee asha maa Miata tat cage SV tokens achat ieee ere! ' Sademaeicemet al rest Aalst Caaeircny lr sete i ake alpen coc ‘oes ave the ep et you ga the Ronda orb tery abet ‘aeetegurncy. Hy ety thay. sd rl ial Sy ecient Eaten fom a alg Aad Seeking =I is voi Seen eaptcn itm nt took andy iSrheasdomarenA oe the ado bcty, OO ak Se cowihir ere SE poodle eve pt lel rad aire chee F FV ne Wp iat ya? Tose eo arco cet (tepid (APD smrcos conto ai VA orcs ae pri ene ‘Sap ln o t owen sty comet i ol fete tg Kew be Gece Aalst ‘ heen eral met ented Vey ete Wry We dacs that evade ot (ane se ove npn toc wane essere you fave’ le Ewa ie wth ah for enpropem ytomerteay (odang seaiedthewt VI conv wie step he wen user neue ‘Se pues the big em or buying "Volto equcsey comers sav uot dang yuna tat you get ice versa ben ee a ae Iereespmmce noes deco 1 yee Plick VF Caner devetlng tat taker # wk, And te cane aes icky rouse VE essa the pesowenk emerson elt el nem waye- The one anon Doni he frid.taieus ‘One centr ibe VIF and WWTP oc /V sure ty ENV ace Raed bran A nner ne poe Ov gon dente ‘oul giee ere, rnbetwe Enew youtove «probly pou ealy vernon d UF convene Bat vary ugh to get ur Apgheaton Noten ipa ae lth angst ees. Ate tp sens ayy at ie sna tha arg curable (617) 329-1610. Or wrtem: Desbu, "Wives stadia we've Manachines 12026 In Eaope. tre kang ht of pac ete Tel 739958, Teer 25881. Or win, some of he pecal ee tring rode 70 eae, Balu ‘pt taeda Like scraper one ‘elder TC on ard the up {0 WORSE el wed il ke that Weregethe 4701200 VOkHL VIF the 4703-— 100 VIE se thet gen — te 4705. IMF VF Once we nated the Vi oe i nt FV we ey. So we've gt the 4702 TUF nl the 4708 FOOLS F/V,neering world. all the trouble of writing about “Yinear” and ana- log circuits? Everybody knows that linear cireuits are dead, Nobody's buying or designing in lin ear circuits; they are all being re- placed by digital signal processors. Analog computers have been dead for years. Why bother? Well, these days, even though there are trends to perform a lot of fanetions with digital computations, people are finding that there are still ahuge number of things that cannot be done properly without analog circuits, IVs true that some of the trendy new radios, claim to use a lot of digital tech- niques, but even there, the receiv- ers and amplifiers, are analog. cir- BOBPEASE —cuits—even if the OBTAINEDA the receiver's fre- BSEEFROMMIF — queney appears to INIANDIS be digitally con- STAFFSCIENTIST trolled. AT NATIONA When people SEMICONDUC. are designing dig- ‘TORCOR?, ital computers, SANTACLARA, they need analog CALE. techniques 0 make good lay- outs for fast buses. They need power supplies— either linear ICs or switeli-mode cireuits (which use ana- Jog circuits internally), And, as for us analog designers, the old-timers and the rookie engineers—well— this column is intended as a soapbox for 86 Fete Tek © NTC |SePreeR 18 29 PEASE PORRIDGE WHAT'S ALL THIS ANALOG STUFF, ANYHOW? This is the first of a series of columns about analog and “linear circuits written by Bob Pease, Staff Scientist at National Semicon- ductor Corp, Santa Clara Calif. We think our readers will get a lot out of Bob's seemingly off-the-wall, yet insightful views of the eng hy? Why am I going to | me to talk about linear cireuits, and then forme to listen to your opinions and comments and questions Thave a lot of opinions, but I'm also very interested in what makes you tick, I may not be the smartest engineer in the whole analog jungle, bat I have sort of vofunteered to start writing this, and we'll see what jnappens—what interesting debates wwe get into. I have a bunch of opin- ions about ICs, data sheets, testing, computer simulation, education, troubleshooting, along with a whole slew of little topics. Inevery darned issue of Blectron- ic Design, I'l try to have some pro- voeative or insightful topic. Some will be pretty technical, others will bbe more philosophieal in nature, But ‘one thing's for sure, I'll try not to bore you. For example: What's all this heuristie stuff, anyhow? HEURISTICS? ‘The other day I was talking with a young college graduate from a pres- tigious Eastern engineering school He explained that his specialty was analog synthesis. I perked up my ears—I hadn’t heard much about that. Where could I read more about this? “Oh,” he said, “in some of the IEEE journals.” Hmm. He started to explain the approach, It's a heuristic approach, he said, Hmm. What's a heuristie? He said, “You don't know what a heuristic is? Really?” I ex- plained no, that we didn’t have any heuristies when I was in school (Note: Mr. Webster says that heu- ristie means “serving to guide, dis- cover, or reveal; specif.: valuable for stimulating or conducting empirical DESIGN research but unproved or incapable of proof—often used of arguments, methods, or construets that assume orpostulate what remains to be prov- ‘en or that leads @ person to find out for himself —from the Greek, heur- iskein, to discover, find,”}—Gee, that sounds like analysis or optimization to me—not synthesis. ‘The young man explained that when you make a lot of optimization experiments, heuristie refers to the starting place, the initial guess. Hmm. He said, “You feed in some requirements and some specifica- tions, and it optimizes the perfor- mance.” Hmm. Now, what circuit does it use? “Oh, it uses the eireuit that you giveit.” mm, Tue Key Question Tf you give ita cirenit that doesn't, work well enough, how doesit gener- ate acireuit that works better’ “Oh, it doesn’t.” I explained to this young. fellow, that in our whole product line, about 99% of the circuits are not optimized at all—at least not “opti- mized” in the sense he understands, If you really OPTIMIZED them, they would all be a little different: than they are now. But each one has a different circuit that is a revolu- tionaty—not just an evolutionary— change from any previous cireuit. So there may be places in our company where optimization is useful and a good idea, But I wish he wouldn't eall it “ana- log synthesis,” that seems to be a misnomer. The circuits around our area—the ones in the NSC Linear data books (and, [bet, in the PMZand Analog Devices data books, too), were not “synthesized” except by bright engineers who knew that the oldcircuits wouldn’t cut it, andanew circuit was needed. Good luck, young fellow! All for now. / Comments invited! / RAP/ Robert A. Pease / Engineer ADDRESS: Mail Stop 25008, ‘National Semiconductor P.O. Box 58090 Santa Clara, CA 95052-8090PEASE PORRIDGE WHAT'S ALL THIS NOISE STUFF, ANYHOW? (PART 1) cently I was invited to a meeting to see the results of ‘2 new high-performance, low-noise transistor project. Tlooked at the technieal report. The new transistors were indeed quieter than the oldones. In fact, they were2 to 4 orders of magnitude quieter han, he conventional ones. T was ious, What was the test meth- od?" Ob, here isthe test cirutt (sce the figure). Now, I asked, were the betas high or low! I was told they were pretty fow, but they can be brought up lat- cr. I explained that when the betas get low, if they're not very well matched, the test cireuit's outputean peg—right up to the + or - rail Then, of course, the apparent output noise gets rather small. (Ohhh!) When this cir- cuit is running okay, the current noise is amplified by a big resis- tance: 1 MO x (N+D, where N is the closed loop q gain, (Rp/R,). So, the output will in Wd clude: (I+ noise of BOB PEASE Qua) + (noise of OBTAINEDA Qs) 1000 Ma. BSEEFROMMIT However, the off- IN1961ANDIS__ set current (I+) — starrscievnst (I-) will also be ATNATIONAL magnified by 1000 SEMICONDUC. _M@. So even 9 nA TORCORP, of offset current SANTACLARA, will cause the op CALE amp's ontput to try to go to +9 V. Ifthe power supply won't let the out- put get to a fair balanee, the output will peg. Naturally the output be- comes very quiet—the circuit. has stopped amplifying the noise. Tako pointed out that ideally the cireuit could measure the base-cur rent noise of the transistor, or device under test (DUT). But the layout of the circuit is quite critical. Just 1 pF of capacitance (C5) from the output to the base of Qyp will cause a lag in the response: 1000 M0 x 1 p= Ims, othe noise will roll off above 160 Hz. ‘Youcan makea layout with less than 0.1pP, but you have to think aboutit and engineer, ‘When we checked the test box, it was laid out very neatly: The output wire was bused alongside the sum- ming-point (base of Qs) wire, and the bandwidth was indeed less than 100 Hz. Ina future column, Iwill talk about what picofarad looks like and the harm it can do to you. So, even if the output wasn't pegged and you looked for the noise at 1 or 10 kHz, this testcireuit would giveananswer that’s considerably quieter than the theoretical minimum for the transis tor. Now here's a good place for sanity check, It's not impossible to measure the noise of a transistor’s base eurrent, but you must have a suitable eireui I wrote a paper back in 1968, and as I look at it today, the only things that changed are the names of the op amps. You ean’t buy any of those old discrete-transistor, potted-module op amps any more, but the testing -But Gewnee Se Worse Resuirs AS LONG AS ELECT TRANSISTOR cuResMr wase TESTER = Vout = ZByse (Qin usa) xf0liMn ae (Zeon-Tedre) s1000H-0. GUS Vos wf000 az Vore xit¥0 MER ee ton teene BLE Yam op 2 0 FRE UT PAT VOLTAGE GOES possiacey WcASE THE ns Liny7d WIRES Make, ALL fi Fea RON TC Qefatbeitesiod| 141PEASE PORRIDGE approaches are just as valid. Maybe Vil write an updated version. fn this ease, the problem was that a LMS resistor and a noise gain of 100 or 1000 (provided by Ry and R,) wasn'ta good idea. The stray capacitances and the noise of the 1-M0 resistor are detrimental to accuracy. It’s bet- ter to use a real 100-MA or 1000-40 resistor. Infact, 2 20-MQor5-MOresistoris, justified ‘because it will still give plenty of signal-to-noise ratio, and a Jot more bandwidth. More on how to do this in the next issue. All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS: Mail Stop C2500A, National Semiconductor P.O. Box 58090 Santa Clara, CA 95062 3090 Make Tracks... 10 your aearest mailbox and Send forthe latest copy of the free Consumer Information Catalog Ielsts about 200 free orlow cost government, publications on copies ike health, nutrition, careers, money management, and federal benefits Just Send your name and address to: Consumer laformation Center Department WT Pueblo, Colorado 81009 US General Serve amlstation (CODE Sten Selection Caray sed ene ‘pote eel (Qa aucoswrrcn’ ‘Get ination ‘Toupee, anions Tnaueil Conroe caRCLE 194 ELECTR onre pes Stores toil 145October 25, 1990 What's All This Neatness Stuff, Anyhow? October 25, 1990 12:00 AM Electronic Design Bob Pease ‘Once upon a time, there was a rapidly converging conflict: My boss thought my office was getting messier and messier, and he wanted me to make it neater. Now, this was just a year or so after my desk had won a $500 prize for being the "Ugliest Desk in Northern California.” So I guess he thought he was justified in pressuring me to clean up my act. He solved that problem by making it one of my goals to get my office to an acceptable (whatever that meant) level of neatness. Well, we never found out what that meant. Every time he would ask me how I was coming on the neatness ‘campaign, I would tell him all of the other things I was doing to help our customers. What ifI came in on a Saturday with good intentions of neatening up some of my office, and the phone rang. Should 1 tell the customer, "No, T won't help you, Ihave more important things to do”? So every year he would mark me down points in my review for not fulfilling my goals. He finally got so discouraged that he left the company. The poor guy. He just wasn't devious enough! He could have waited until the next earthquake and told one of the guys to knock over a couple of my piles of papers. Then he could then explain that I had to get it at least toa reasonable level of safety. But he never figured that out, and I didn't tell him until after he left Some people keep their desk neat because that’s what feels good to them. [find that neatness is not a priority compared toa number of other things, such as answering the phone when a guy needs help, or volunteering advice when a ‘customer has a problem. Some people find it easy to keep a neat desk because they throw out things that make it look ‘messy. I just don't operate that way. (One time I was working on a Saturday after being at National just a few months. My desk was already stack up pretty hhigh. Another guy was at his desk, which had just a few dozen things on it. He was picking them up, one by one, ‘studying them, and then throwing most of them in the wastebasket. I commented, "You sure do keep your desk neat.” find something I don’t need, I throw it out.” I said, "Doesn't your wife ever get nervous?” He replied, "It’s my third wife ..” No, I don't operate that way. ‘One day, an engineer stepped gingerly into the entry way of my office and asked, "Bob, do you have a Siliconix catalog?” ied, "Sean .. you're standing on it.” He looked down and, indeed, he was. He was impressed. But I knew right it was, because I had recently tossed it over by the doorway so I could then put it in the bookease by the door. Sean just happened to walk in before I put itin the bookease. More recently, I inherited a couple of filing cabinets and a huge 7 ft.x 3 ft.7 ft. cabinet from a Fairchild laboratory. ‘Our secretary explained that I would have to junk it unless I could find a use for it. Tsaid, "Well, Teould always put it in my office.” She looked at this huge ark and said, aghast, "No, you couldn't do that.” thought about it.I got a yardstick, and I figured out that, with an ineh to spare, I could do that. My technician and 1 ‘spent nearly all morning reassembiing that cabinet and easing it into the corner of my office. I put about 1/3 of a millior ‘cubic inches of my paperwork into that, and into the other fle cabinets, and improved the appearance of my office so ‘much that our senior secretary admitted that I qualified for an "Enviros Award.” In the past, the various departments ‘would vie to achieve cleaner clean rooms and higher-yield fab lines by having better cleanliness. A whole department of 20 or 30 people would work real hard to cut down the number of particles in their area and win an Enviros Award. But got my Enviros Award single-handedly. T hate to guess how many particles I straightened up. Right now, my office seems to be in the getting-messier-again phase. When T have to review a mask set, with precision down to the last tenth of a micron, I get my head in the right mood to do that. And when I'm done, in sheer rebellion I guess, I abandon the neatness for a while. I save what seems to me to be of value. Often that includes documents and papers and notes that other people would think aren't very valuable - until they come to see me years later, hoping I ‘ight have the information they need. Often I do. Go ahead, call me retentive. See if T care. Now that the NBS has changed its name to the "NIST" or "National Institute of Standards and Technology,” have figured out the next way to enhance the neatness of my office. I'm going to buy a big dresser with 6 big drawers and a mirror and everything. I'm going to put it right at the entrance of my office, and put our ultra-precision resistors and ‘capacitors in those dresser drawers, And I'm going to call it "The National Bureau of Standards.” Comments invited! / RAP Robert A. Pease / Engineer RAP Update: This was the fourth column I wrote, the fourth to be published. And suddenly I began to get some Fan Mail. Alot of people said they had desks that were pretty close behind mine in sheer messiness, My desk still is a Federal Disaster Area. I tried to put a recent picture of my desk in this web page, but it isn't even recognizable as a desk. Isit?PEASE PORRIDGE WHAT'S ALL THIS CMRR STUFF, ANYHOW? cently, many people have asked me about how to test ‘op amps for common-mode rejection ratio (CMRR), which is defined as the delta of the offset voltage (Vos) versus the com- mon-mode voltage (Vos. The first thing T tell them is how not to mea- sure OMRR (Fig. 1). If you drive a sine wave or triangle wave into point A, it seems like the output error, as seen by @ floating scope, will’ be (N+U) times (Voy divided by the MRR). But that’s not quite true: you will see (N++1) times (the CM error plus the gain error). So, at moderate fre- quencies where the gainis rolling off and the CMRR is still high, you will see mostly the gain error, and your curve of CMRR vs. frequency will Took justas bad as the Bode plot. That’s because with this cireuit, that’s just what -you willbe seeing! Tecarns outthat afew opamp data sheets till exist in which the CMRR curve is stated to N BORPEASE —ethesameas the OBTAINEDA ——_Bodeplot. The Na- BSERFROMMIT tional LF400 and INI961ANDIS_ LFM0L are two ex- STAFFSCIENTIST amples; next year ATNATIONAL — we will correct SEMICONDUC. those curves to ‘TORCORP,, show that the SANTACLARA, common-mode re- CALIF, jection ratio is ac- tually much high: er than the gain at 100 or 1000 Hz. Ah, let’s avoid that floating scope. We'll drive the sine wave generator {nto the mid-point of the power sup- ply, and ground the scope and ground point & (Fig. 2). Then we'll get the true CMRR, because the out- put won't have to swing. Right? Wrong! The circuit funetion hasn't changed at all; only the viewpoint of the observer changed. The output does have to swing, referred to any power supply, so this still gives the same wrong answer. You may say ‘that you asked for the CMRR as a funetion of frequeney—but the an- swer is, in most eases, the curve of gain vs. frequeney. What about, as an alternative, the well-known scheme where an extra servo amplifier closes the loop and doesn’t require the op-amp ontpat to ao any swinging Fig. 9)? That's okay at de. So it’s adequate for de testing with automatic test equip- ment (ATE), for production test, and for stepped delevels -Andit wiligive thesameansweras ‘my cireuit (which P'll discuss a little later) at all low frequencies up to where it doesn’t give the same an- swer. Now what frequency would worry about, I'l just avoid that by using a cireuit which gives very con- sistentand predictable results. Wher I ran an LF856 inthe eireuit of figure 1, Igot an error of 4mV ple pkat 1 kHz—a big fat quadrature er- ror, 90 degrees out of phase with the output (Fig. 4, upper trace). If you think that’s the CM error, you might say the CMRR js as low a 5000 at 1 kHz, and falling rapidly as the fre- queney increases. But the actual CMRR is about 0.2 mV pl-pk (Fig. 4, lower trace). As a result, the CMRR isabout 100,000 at1 kHz or any lower frequency. In addition, the CM error on this unitisn'treally inear. As you get near ~9 V, the error gets more nonlinear (this is a-9-V to ¥12-V CM range on a 12-V supply; I chose a £12-V supply so my function gener- ator could overdrive the inputs). ‘As you can see, this business of CMRR testing isn’t trivial Just how, then, can we test for CMRR and get the right results? Well, there’s a ‘darned fine cireuit I invented myself about 22 years ago (Fig. 5). It has limitations, but it’s the best eirewit T’ve seen, Let's choose Ry= Ry, = 1k, Ry = Rip= 10k, and Ry= 200k and Ry ‘8 5000 single-turn carbon pot, oF its equivalent. ‘These values will permit us to set. up 2 moreor-tess halaneed bridge, that be? Nobody knows! Because if | witha fine trim for de balance. In this you have an op amp with low CMRR, the servo scheme work accurate- ly up to one fre queney. And if you have an op amp with high CMRR, the servo scheme will work accurate. iy only up to a dif ferent frequeney. Also, the servo amplitier adds so much gain into the Joop that ringing, overshoot, or mar- ginal stability at some mid frequen- cies is inevitable. That's much too horrible for me to ELECTRONIC DESie Ny |107 "NOVEMBER 1990PEASE PORRIDGE cr Eine enree cewe.n, 2) FIGURE 2: case, the noise gain is defined as R/ R,, + 1),orabout 11. Let's puta +11- voitsine wave into the signal inputso that the CM voltage is about +10 volts. The output error signal will be about 11 times the error voltage plus some function of the mismatch of all those resistors, Okay, first connect Vj, to the seope’s horizontal input, Vou to the vertical input, and operate the seope in cross-plot (X-Y) mode. Trim pot R, until the output error is very small, or until the slope is nominally flat ‘We don’t know if the CMRR error is balaneed out by the resistor ertor or whatever, but it turns owt we don’t care. Just observe that the output er- ror, as viewed on a eross-plot scope, is quite small, Now connect in Ry, 2 nice low value, such as 200 0. If you Geo.) Flours 3 TI CMRR. fe Femrenance 108 lRotiatiots tor 9 ' © ureer Enon. Glee) x Ott) wee gb NEE Route SH THS 18 “STA wor A GD TEST cer oer mrt 6 AvequTe DC TEST cima ON AuL me Re SCs Cus gine ‘sit down and com- pute it, the noise gain rises from 1) to 111. Namely, the noise gain was (1+ R,/Ry), inereasing to (+ Ry/R) plus (By + Ryg/Rigy In this example, that’s an increase of 100. So, you're now looking ata change of Vag: eaual to 100 times the input er ror voltage, (and that is Vox divided by CMRR), Of course, it’s un- likely for this error voltage to be a lin- ear function of Vey, which is why I recommend you look at it with a seope in X-Y mode. Too many people are inclined to make a pretend game that CMRR is constant at all levels and CM error is a linear function of Voy. So they just look at 2 points and assume every other voltage has lin- ear error; and that’s just foo silly. Even if you want to use some ATE ‘you will want to look at this error in ‘atleast places—-maybe even at 4 or Svoltages. Another good reason to use a seope in the X-Y mote is so you can use your eyeball to subtract out the noise, You certainly ean’t use an ac voltmeter to detect the CMRR error. For example, with the waveform of figure (lower trace), the CM error is fairly stated as 0.2 mV pk-pk, not 0.5 mV pkepk (as your meter might say if you let it include the noise) Anyhow, if you have a good ampli- fier with a CMRR of about 100 4B, the CM error will be about 200 nV pk-pk. When it’s magni fied by 100, you can easily see an output error of 20 mV pk- pk, If you have really good unit, with CMRR of 120 or 140 dB, you'll woup OFF ered 1) pap zerit® DESIGN ‘want to clip in the Ryyau such a3 20.9, and then the delta (noise gain) will be 1000. The noise will be magnified by 1000, but so will the error, and you can see what you need to see. Now, I won't get embroiled in the question: Are you trying to see exactly how good the CMRR really is, or just if the OMRR is better than the data- sheet value? In either ease, this ap- proach is the best way I have seen. For use with ATE, you don't have to look with a scope; you can use a step or trapezoidal wave and look justatthe delevelsatthe ends, or the middle, or wherever you need. Note that you needn't trim that resistor network all of the time, nor do you have to trim tperfectly. Allyou have to know is that when the noise gain changes from a low value to a high value, and the output error changes, the change of the output error is of interest—not really the pk-pk value before or after, but the delta, You dori’ have to trim the resistor to get the slope perfect, but that’s the easy way forthe guy working this bench to see and appreciate the changes, This is a great cireuit to fool around with. When you get it run- ning, you'll want to test every op amp’ in your area, beeause it gives yousucha neat high-resolution view. It gives you a good feel for what's happening, rather than just hard, cold, dumb numbers. For example, it you see a 22-aV pk-pk output signal that’s caused by a22 u¥ error signal, you know that the CMRR really is ‘way up near a million, which is much more educational than a cold “119.2 aB" statement. Besides, you learn rather quickly that the display’s slope and curva- ture are important. Not all amplifi- ers with the same “1192 dB” of CMRR are actually the same. Some havea positive slope, some may have a negative slope, and some curve madly, so that if you took a 2-point measurement, the slope would change wildly, depending on which two points you choose (if you in- crease the amplitude of the inputsig- nal, you can also see plainly where severe distortion sets in—that’s the extent of the common-mode range). Limitations: If you set the noiseieee Li Wa CS PICO’s Ultra-Miniature Power inductors are ideal for Noise, Spike and Power Fil- tering Applications in Power Supplies, DC-DC Converters and Switching Regulators. © OPL standards available MIL-T-27/356 © Temperature range —55°C to +130°C © All units are magnetically shielded © All units meet the requirements of MIL-T-27 (TF5SO4ZZ) ‘© Minimum possible size © Split windings ‘© Inductance values to 20mH with DC currents to.23 amps CO manatactures complete lines of Delivery— stock to Electronics, Inc. “ESN aceon Py. ME ero 108, Call Toll Free 800-431-1064 new york can, 914-699-5514 OUR 110 lRotintisia ino” ° S ' PEASE PORRIDGE Figures gain as high as 100, then this circuit, of course, willbe 34B down at (Few) divided by 100, 50 you would only use this up to about 1 kHz on an ordinary 1-MHz op amp, and only up to 100 Hz ‘ata gain of 1000, That's not too bad, really. Tolookat CMRR above 1kHz, you might use Rage = 2k to give good re- sults up to 10 kHz, In other words, you have to engineer this cireuita lit Ue to know where it gives valid data. You can’t avoid the fact that think- ing is required. Sorry about that. For really fast work, I go toa high- speed low-gain version where Ry Ry =5k, Ry— Ry 5k, and Ri = 2k or tk or 0.5k. This works pretty well up to 50 kHz or more, depending on what gain-band- width produet your amplifier has Por best results ata, it’s important to avoid stray ca- pacitanee of wires or of a real switch at the points where you connect to Rio, for Rigg Usually T get excellent re- sults from jast grabbing on to the resistor witha mini- gator clip. You ean avoid the stray pF that way. If you use a good selector switeh, with all of the wires dressed neatly in the air (which is an excel- lent insaiator) you may be able to get Yost Rr = Re Ra = Re DESIGN ae alt Vem = Vare @xfleses) eemt ~Veascmne Vor Vewece Crees + Resto decent bandwidth, but you should be aware that youre probably measur- ing the ac CMRR of your set-up, not of the op amp. Actually, I was cussing this circuit with a colleague, when I realized the best way to make up this 20-0 resistor is to connect one 10-1 resistor to each input, and then clip the other ends together with a mini-gator clip up in the air. Bal anced strays, and all that. If you have an op amp with low gain or low gm, you may want to add in a buffer follower at ab (Fig. 5, ‘again),so the amplifier does not ge erate a big error due to its low gain. ‘The LM636] would need a buffer as it omly has a gain of 3000 with a load of 10k, and its CMRR is a lot higher than 3000, Altogether, I find this cir cuit has better resolution and gives less trouble than any other circuit for measuring CM error. And the price is right: a few resistors and a mini-gator clip. All for now. / Comments invited! / RAP/ Robert A. Pease / Engineer ADDRESS: ‘Mail Stop C2500A, National Semiconductor P.O. Box 58090 Santa Clara, CA 95052-8090 FIGURE 5: - A Good TEST CRENT FER CARE use Mice re Sexter Rio eLer 3 = 40*Rt Ras Rafoo Aus CRensros CRBSPEASE PORRIDGE WHAT'S ALL THIS SPICEY STUFF, ANYHOW? (PART 1) ecently, I was down in New Orleans at one of the IEEE conferences—the interna- tional Symposium on Cir cuits And Systems. The keynote speaker, Professor Ron Rohrer from Carnegie-Mellon University, com- mented thoughtfully about many as- peets of education for engineers. But what he said that really stunned me ‘was his observation that “In the era ‘f Spice, nobody designs on the back of envelopes any more”. Ouch! It is becoming more and more true that young (or lazy?) engineers can’t do ‘uel designing without some com- puters or high-powered caleulators, They really cannot design things without alot of assistance from com- puters. Spice just happens to be one of my pet peeves, and I will stare gnawing on its an Kes today. Now, I've al- ways "been a friend of analo- gies, analogues, anaiogs, similes, models, and meta- « phors. When I worked at George \ A. Philbriek Re- BOBPEASE searches, the com- OBTAINEDA —pany’s motto BSEEFROMMIT was,"“The analog INIG6LANDIS way’ is the model STAFFSCIENTIST way.” In those ATNATIONAL days, we sold SEMICONDUC. some analogue TORCORP,, computers, even SANTACLARA, though that part CALE, of the business was shrinking and the popularity of the op amp was on the rise. But we al tried to follow the party line: that analog computation vwas@serious business. And tstilis, although as a pereentage of the elee- tronies business, it has shrunk to a ELECTRONIC tiny fraction. Still there are many times where a little analog computation is exactly the right thing, and someday I will expound on that... But, Spice (Simu- lation Progeam with Integrated Cir- uit Emphasis) is a rather popular and powerfu tao! these days, and al- most everybody finds it useful to some extent. I remember when my ‘old boss, Tim Isbell, showed me how to use it~and then we spent half a day horsing around beeause it said we had 72 forward voltage across a diode, but there was no current through the diode. I will emphasize justa few of the basie problems with Spice today. ‘The main problem is that people tend to trust its answers, as they trast most computers, long after the reason to trust it should have evapo- rated. I have come very close to fist fights and screaming contests when a person claims that such-andsueh an answer is obviously rightheeause Spiee gave it to him. Conversely, I normally try to avoid working: with Spice unless I ean first run a calibra- tion program on it, soit gives me an answer that makes sense—a sanity cheek This is mach fike the old days of the slide rule: You couldn’t use the slide rule unless you already knew approximately what the answer was, It's not like a caleulator, where the decimal place is provided on a plat- ter. You have ta provide your own decimal place. In other words, you fare forced to be a pretty good engi neer before you even pick up your slide rule, or your analog eompiter, Bat people who use Spice are often buffaloed or fooled by any absurd kind of answer. So, trusting your computer seems to be one of the new trends that I would like to see quashed. It’s too DESIG ‘NOVEMBER 22.190 easy to find, weeks later, that the computer told you a lie, beeause the data you entered had a typo error or amonumental goof. Now, never letit be said that Pease recommends you use analog eomput- ers or breadboards instead of Spice because analog computers don’t make errors. Spice lies, but analog computers do not? Oh, please, don’t say that: analog eomputers lie, too, and so do breadboards. But I prefer them because they often offer a greater insight and understanding as to what's really going on. Soif you survive their problems, you are smart enough to keep out of other kkinds of trouble. But that’s just a bunch of philosophical stuff. ‘The thing that makes me nervous about Spice is that it was largely de- signed by 2 group of grad students (Laurence Nagel and others at Berkeley) back in 1978, Now, when ‘you find a problem, a discrepancy, a Blitch, a flaw, an ervor that seems to be builtinto Spice, can you go back to the people who designed it? Hardly, There's no continuity. There are some people who claim to ‘support” Spice, but I'm not usually impressed with their statements, My biggest gripe with Spice is its lack of convergence. The ordinary Spice 266 has all sorts of problems, even if you don't use FETs (we find that FETS usually make the eonver- genee situation really unhappy). For example, one time I had a moderate- sized eireuit with about 39 bipolar transistors, and it didn’t converge well. Then, all of a sudden, one day it started to converge beautifully and quickly. vas so impressed, I backed up to find the"scene of the crime.” J tried to duplicate ail of the changes I had made since I last had problems. It finally turned out that [had an unused resistor and an unused ca- pacitor each tied from one node to ground. Nothing else was connected to that point. They were originally “commented out” by an asterisk. But, atone point, I deleted the aster- isk and the useless Rand C were dropped into the eireuit—they just happened tomate the convergence NULTRA-MINIATURE DC-DC Converter Transformers Electronics, Inc. ‘538 acQvesan Poy Vera. 10882 Call Toll Free 800-431-1064 new York cat 914-699-5514 ‘Send for PICO's new catalog fexturing Uta Moiature Transtormersiinductors/ De-DC Converters cRCLE I Blevectronic PEASE PORRIDGE lotbetter. When I removed the R and C, things got worse again. This lead me to appreciate two things: that the convergence is a lot more fragile than we suspect; and that we may be able torandomly throw useless resis- tors into a cireuit, and sometimes they may help to improve the conver- gene. In other words, if you have a cir cuit thatshows bad convergence, the computer mighthavea subroutine to randomly sprinkle a few resistors into the eireuit and see if that helps— akindof “autoconverge” scheme. At present, we're still working on this, and it may bea useful approach, But, this concept isn’t too surprising if you've ever heard that the conver- gence of acireuit may beimproved or degraded depending on the names and numbers you call the nodes of the circuit. Tf you swap a couple nodes’ num- bers, and things get better (or worse), doesn't that make you ner- vous? Orat least suspicious? Some people claim that dynamic ramping up of the power supplies can help improve the convergence. Maybe. Lagree, it's worth atry, butT recall a lot of cireuits where even ramping didn't work. Another seri- ous problem I had with Spice was when T ran some simple transient tests—triangle waves—on the col- lector of a transistor. I ramped the Vo up and down from 5 V to 15 V, back and forth, and ran several tests ‘Then I added some complicating fac- tors, and wanted to look at the eireuit for the first 202 us of a 1-H trian- gle wave. ‘Namely, after the first two eyeles of the triangle wave, I decided to look at the collector current (i= Cx dv/dt) of the transistor at t= 201 8. I got my answer printed and plotted, and it didn’t make any sense. I stud- ied the whole circuit, and I used ev- ery troubleshooting technique I could think of, and it didn’t make any sense. The current through the 1 pF of Cy was not 0.2 4A, but 5. How could that be? ‘After several hours, I finally de- cided to look at the incoming wave- form (which [knew would be a waste NOVEMBER, 1990 of time because I knew very well what I had told it to do... !had com- manded it to go back and forth from 5 V to 15 V, at the rate of 50 us per each ramp). But when I looked at 201 ws, the dv/dt had suddenly in- creased from 0.2 V/s to 5 V/us. It turned out that because I had com- manded the Plot mode to stop at 202 us, the transient generator had de- cided to go from 15 V to 5 V—not in the time from 200 to 250 us, butin the span from 200 to 202 us. The dv/dt sped up by a factor of 25, without be- ing asked to, for a eompletely unex- pected reason, Nothing I had ever seen about Spice, nothing my friends had ever heard, would lead you to expect this. In faet, Spice sort of encourages you to look at the waveforms any time you want—it offers a sort of “infi- nitely versatile, expanded-seale os- cilloseope,” and if it has a dv/dt that suddenly changes, well that’s quitea surprise, So, I immediately wrote an open message to all of my friends at Na- tional, warning them about this po- tential problem; and now I'm writing about this to warn all of my friends everywhere. These are just some of the reasons I’m not, enthusiastic about Spice. It’s goofed me up, me ‘and my friends, too many times. My boss points out, it’s not neces- sarily true that all kinds of Spice have such bad problems with eonver- gence or bad computations or spuri- ‘ous signals. And that may be so. If somebody who knows all about dif- ferent Spice produets wants to write in and assure our readers that his Spice will never do that, well, that’s fine by me, But, meanwhile, just re- member—I don’t hate these digital computers. They hate me; I despise them. Stay tuned for the next col- umn, because I’m not finished with Spice yet. All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS: Mail Stop C2500 National Semiconductor P.O, Box 58090 Santa Clara, CA 950528090 DESIGNPEASE PORRIDGE WHAT'S ALL THIS SPICEY STUFF, ANYHOW? (PART 11) he other day T was standing out in the rain, talking witha design engineer from the East Coast, He saidallof the other engineers at his company ridi- cale him because they rely on Spice, and he depends on the breadboards he builds, There's just one hiteh: his circuits work the first time and their cireuits don’t ‘To add insult to injury, his boss forces him to help his colleagues get their cireuits working, since he has 80 much time left over. I said that sounds pretty good to me, so long as his boss remembers who is able to get out the cireuits witen it comes to doing reviews for all of the guys. ‘This guy gave me a tip: Don't de- signa circuit in Spice with 50-0 resis- tors. Use 50.1 0, it converges better. Hmmmm. That sounds kind of in- triguing. Right now I'm struggling with a Spice model of a circuit. Not of a new cireuit, but of an old eireuit: the bandgap refer- ence of the old BOBPEASE —LMS31 that I put OBTAINEDA into production BSEE AROMAT back in ’77. It's a IN1961 ANDIS good thing I putit STAFFSCIENTIST into produetion ATNATIONAL — before we got SEMICONDUC Spice, because if I TORCORP, had first run this SANTACLARA, through Spice, 'd CALE, have been pretty ——— iiscouraged. Spice says this eireuit not only has a rotten tempeo, but that it oscillates like politician T went back and double-checked the actual silicon cireuits. They soar like an angel, have very low tempeo, BLEcTRONIC and are deat-beat when you bang an them. They haveno tentieney tooscil- late; they don't even ring. So why does Spice persist in lying to me? Doesn’titrealize I will break its back for the impertinence of lying to the Cear of Bandgaps? Vm alittle busy right now, but in a while I will find ‘out why it lies to me. The Spiee and CAD experts around here tel! me, “Oh, you must have bad models.” I've been told that before, when T ‘was right and the experts were abso- Jutely wrong (I mean, how can a sin- gle FET oscillate at 400 kHz?” With the help of 2 resistors... More on this topic later. Tye already gotten several letters from readers expressing general in- terest and enthusiasm coneerning this column about linear circuits. Al ready a couple writers have asked, “How about all of these new models for op amps? Won't they lead linear designers in anew direction?” My re- plies to them start out by covering couple examples of old op-amp ma- eromodels that have raised ques- tions for over a dozen years. A guy calls up and asks me, “What is the maximum de voltage gain on an LM108?”' Ireply, “well, it’s 40,000 snin,, uta lot of them ran 300,000 or 500,000, and some of them are as high as'3 or 4 million.” The customer sighs, “Oh, that’s terrible...” When I ask why it’s terrible, he explains that when the gain gets high, the gain- bandwidth (GBW) produet will get so high that it will be impossible to make a stable loop if the GBW prod: uct gets up to dozens or hundreds of megahertz. Sigh. I sit down and explain thet there's no correlation between the de gain and its spread when compared to the GBW product and its spread, ‘The guy says, “Oh, I read in a book somewhere that there's good corre- C DESIG DECEMBER 1,190 | lation, because the first pole is con stant.” I tell him to throw out the book, or at least X out those pages, because the first pole is nofat acon: stant frequency. ‘These days, l read that several op- ‘amp companies are giving away free Spice models. What do I think of these models? Well, on a typical basis, 1 have read that some are pretty good. In several typical situations, they slew and settle (and ring just a litle, as real op amps do) and have as good accuracy as a real typical op amp and its feedback resistors. Maybe in afew years, models of slow op amps willbe trustworthy. But I don't think you can get very good results from modeling the fast ones. Why? Pe- board layout strays, Bnough said, ‘And besides, how good are those models if you ask their makers? Are the models guaranteed to givesueha good representation of reality thatif Spice gives good results, the op amps are guaranteed to work? Well.no, not exactly. In fact, from what I've read, none of the op-amp models are guaranteed for anything. The only thing they cando“guaranteeably”’is, give a customer something when he begs for Spice models. It's guaran- teed to make the customer go away happy and to keep him busy for a while. But it’s not guaranteed to ‘make him happy in the long run. This is because the performance of high- speed op amps and precision cireuts depends so critically on the layout and on the resistors and capacitors, making the model itself almost irre. evant, ‘Now some people might say, “How does Pease dare to say that?”* Its easy. T haven't got any Spice models of my op amps to give away. Notatthis time, And if did, or when 1 do, I won’t be able to guarantee them either. Atbest, Imay beable to say, “If you are a good engineer and use these models a8 a too] to pioneer some experiments that are inconve nient to test on the breadboard, you may find these models are helpful But you had hetter cheek things out, | witha breadboard to confirm the cit cuit. For example, you can use Spice soPEASE to ‘measure’ some voltages or cur rents that are so small and delicate that you really could not measure them with a scope, a buffered probe, or current probe—not in the real world. But, if you try to rely solely on these models, without bread- boarding, they won't tell you the whole story. Your erutches will cok lapse, sooner or later, and you can't PORRIDGE capacitor. The same bolds — fgnrt Even the ones that had true fora resistor. only a little time to put in Bat Tean guarantee that think YOU showed that macromodels every opampyoucanbuyor —canbeg _were feasible and effective make has some characteris: gr gtpal gr &54 way todo good analysis ties that ean’t be absolutely while saving computing ‘modelled by any computer DOFFOWA time. ‘That was the primary ‘model. If you happen to de- transistor objective of the study. But pend on that feature, or the fg even the ones that putin the Absence of that featare, i's MOM2l that’s ee ine ac analysis didn't only amatter of time before guaranteed. recognize (or didn't com- say I didn’t warn you.” I showed this eolumn to Bettina Bri in amplifier marketing, and she said, “Bob, you can't say that.” 1 said, “Oh, tell me where I have said anything thats untrae,and I wil fix it.” She admitted that what [had said probably was..quite true, Then T said, “Well, why try to soft-pedal the truth, and pretend that you ean trust computers all of the time? Wouldn't that be a disservice to our custom- ers?” And Bettina replied, “When we have models, we'll have to try to edu- cate our users. We'll point out when you can trust the models, and when ‘you shouldn't. So, after that, are we in disagreement?” Well, maybe we didagree after all At present, we havea smalllibrary of op-amp models released with Analogy, Beaverton, Ore. They're only level 1 models, (low precision), and while we have made some pro- gress on good-precision ones (level ID, they’re not released yet. These are “behavioral models” yather than Spice models, and we think they have several advantages over Spice mod- els. There are some min/typ/max specifications that pretty much cor- respond to data sheet limits. If you use them wisely, they may be help- ful—subject to the conditions Tisted inthe previous paragraph. ‘These models aren’t free, though. ‘They're not even cheap. But you getin trouble. Iwill also guarantee that ment about) that the noise of the reference and the just because you made one bread- | comparator were rather large, and board, and it works well, you can’t put that circuit into production and ‘get 1000 units in & row to work well. Tnless, of course, you've a smart en- gineer and design the circuit “prop- erly” and do your worstease design studies, and plan for well-behaved frequency response, andsoon. And I think that’s true no matter where you buy your op amps, What's new? ‘What color is the king’s new under: wear? Ditty gray, same as every: body else's Tthink there are a number of Elec- tronie Design’s readers out there who will want to comment on this topie, You may be dubious or skepti- calof Spice models. You may be dubi- ous or skeptical about my views. Your comments are invited, You may have experience with Spice or other macromodels. Good? Bad? You tell me, and I'll pass along your | comments to the editors (we may have to allot a little extra space for the Letters-to-the-Editors column fora while). The guys who believe in Spice maeromodels, whether they're somebody that buys or sells op amps, well, they'realsoinvited to write in, I promise to faithfully pass all of the letters along (with appropriate com- ments on the side). But 1 wethinkthey'reworthwhat Why try think you ean already tell youpay forthem,Still,none soff-pedal how skeptical lam. of these models are guaran- Twas at_an evening ses: Now, seriously, where and prete cuits and Technology Meet can you get a model of «you can trust ing in Minneapolis event. transistor that’s guaran- ‘everal companies that sel teed? And to run under aii COMPULEFS — C7"y,c15 had done some conditions? Idon't think you allthe serious work to analyze the canbegorstealorborrowor time? —_cireuit for a 12-bit a-d eon buy a model of a transistor that’s guaranteed. Or of a verter (ELECTRONIC DESIGN, CR 0025,» 16). Mevecrronic DRCEMBER 13,3990 you could not achieve 12-bit resolu- tion without slowing down the re- sponse a lot more than you would have to do otherwise (for a circuit ‘where you didn'thave toconsider the effects of noise). If a good designer of ADCs had these tools, and he knew where to look for noise, or witere to insert lead inductance or extra substrate eapaci- tances, he might use some of these CAD tools tohelp himdesigna better ADC. Batif he just believed what the computer told him, he would proba- bly be badly fooled, ‘Onee, a customer called me up and asked me how to get my LM08s to stop oscillating in his circuit, He ex plained it was a simulated LM108 with some simulated feedback resis- tors, and simulated switches and fil ters. Hmmmm. I asked if he had made up a breadboard, and if it oscil lated, He said he had made it an it didn’t oscillate. Hmmmm. I asked him, “If you built up a breadboard and a computer model, and the real breadboard oscillated, but the com: puter did not, you wouldn't be calling up to complain, would you?” He stopped and thought about it. He cogitated for a while, He said “I'll call you back.” And he hung up. Be never did call back. T mean, what would you do? All for now. / Comments invited! (Now there’s an understatement.) / RAP / Robert A. Pease / Engineer ADDRESS: Mail Stop C2500, National Semiconductor P.O. Box 58090 Santa Clara, CA 95052-8090 DESIGNPEASE PORRIDGE WHAT'S ALL THIS SPLICING STUFF, ANYHOW? everal months ago, a reader wrote in to one of the local newspapers, “If I want, to move my speakers a few feet further from my amplifiers, can I splice in a few more feet of speaker cable, or should I buy all new eable? My brother-in-law claims that splie ing would hamper the sound.” The resident expert at the paper stated that the brother-in-law was wise, as the spliced wire would give inferior audio results. promptly wrote in to the resident ‘expert, asking him on what basis he could Say this. Was he claiming that hhe could hear the difference? T de- tmanded thathe show us readers how the spliced wire could possibly make any difference. I challenged him to listen to any mu- sic, under any atr dia conditions, and I would swap in. various. pieces of speaker wire (enclosed in box es, on a double: blind basis} that had 0 or Lor2or6 or 12spliees, How, short of clairvoy. 508 PEASE ORPEASE ance, could he tell Doternorswr Which wire had Roe CMM the splices, using Shieeseteiner ordinary ‘audio- ATNaTiONAL fFequeney sig. tote AG. nals? OF course, if TORCORP., pede used wal ae edance analyzer SCARS with « bandwidth SE several giga- hertz, you could “see™ some of the splices. But, for good high-fidelity audio, there’s no way you could discern this, especial ly as a splice may make the wire's impedance lower or higher or un- changed. me compelled to comment and raise doubts. There are some experi ments that even I can propose and that I could eonduet, that would be decisive, if the “expert” did not duck the challenge. ‘Now, there are many persons who have golden ears and will claim that they can easily distinguish between vyoliencastangath nee HOM Shortof Pct opeaker eats, er wrote back, So, I'sent Clairvoyance, — However, when these per- Meeak ckepties groups COUGE tes nd et they oualy called “BASIS,” the Bay WHICH WITE have a strong tendaney to Area Skepties Information had the demur. Some people like to St get ie ny, SMBeSUng he ers and in their newsletter, FINALY “this to the tendency of they printed my complaint, audio- cockroaches to seuttle into Shera clams tats RQUENEY i Starnedon the dead, or summon six Signals? Tyas only slighty con- its, or show extrasensory perception, then we must apply some skepticism so as not to encourage gullible persons to invest their money in these hoaxes. But if person who is endorsed as the “high-fidelity expert” says that you can hear the difference between Spliced and unspliced wires, then we, as technical people, have an obliga- tion to express our doubts and our skepticism. Why should a hii sales- man beable tosell a bright-eyed yup- pie a $50 hank of speaker wire, (or $100 or $200 or $400 or more, which is where the really high-end speaker wire is priced these days—believe it ‘or not) just beeause an “expert” says it's beiter to buy new wires rather than splice ona few extra feet? Obvi- ously, ethies in technical electronies ati science is involved here Many hi-fi experts, with their “golden ears,” claim that they can hear differences in sophisticated speakers, expensive amplifiers, or just faney wires, that I can’t possibly discern or detect. Ttmight take many thousands of doitars to just buy the equipment and duplicate the exper- ment, And, their ears might be cor- rect—much more discerning than mine, more than I could imagine Bat, when the “expert” talks about wire and splices, then 1 find myself LEcrRoNte ‘DECEMBER 2, 1900 cerned about how to con- duet the test, beeause ta do a fair test, you might have to change back and forth from, say, speaker wire # 1tospeaker wire #2or #6.If you do that with screwdrivers and pliers, it might take a long time to make the changes; a eritieal listen- er’s judgment might be affected by Jong delays, and it would be unfair to ask for good judgment under those conditions. But if [ proposed to use a number of selector switches, the man with the “golden ears” might argue that the switch's impedance would be worse than the splices, so a switeh would be suspect! No, you can’tuse switches when you wait to doan A-B comparison! But in the last few weeks, the hi-fi review column of this “expert” was discussing how he compares differ- entspeakers: Hesaid to change from ‘one set of speakers to another, he "uses switches! I just hope the switch fs don’t cloud his judgment, as if they were (God forbid) splices. Alt for now. / Comments invited! / RAP/ Robert A. Pease / Engineer ADDRESS: Mail Stop C2500 National Semiconduetor P.O, Box 58090 Santa Clara, CA 95062-8000 DESIGNLaka Supplies © input Voltage 90 to 130 VAC (47/440Hz) © Single, Dual, Triple Outputs © 1200V Rms Isolation © Low Isolation Capacity Avaitebie © Continuous Short Circuit Protection © High Efficiency © Fully Regulated Voltage Outputs © Operating Temperature = 25°C. to +70°C. with No Heat Sink or Electrical Derating Required © Expanded Operating Temperature Available (55°C. to +85°C. ambient) © Optional Environmental Screening Available ‘one week PICO Electronics, inc. 1591 cu Py Mt Veron NY 0 Cali Toll Free800-431-1064 new vone cal, 914-699-5514 Ginter Me Lecrronic PEASE PORRIDGE WHAT'S ALL THIS TESTING STUFF, ANYHOW? Ihe other day, Graham Bas- kerville, my boss’ boss, and T got intoa screaming contest, But we weren't in disagree. ment. We were in complete agree- ment that there was a problem, and ‘we were thinking of examples to il Tustrate the problem. One of the first examples was “The Noise Test.” Many years ago, in @ quarterly report, a manager glowingly explained thaton the mod- elXY7Z, the yield loss due to noise had been cut toa trivially negligible level for the last quarter. Graham said that he inquired politely how this was accomplished. After a little in- vestigation, he found that the “yield” was’ excellent because the test circuit was broken. To restore sanity, he made sure that the noise test now had a lower limit, so if the tester was broken, nothing would pass, and the tester would have to be re- paired. Now, that 8 pared: Now, dt ago, but the gen- BOBPEASE — ral problem of OBTAINEDA ensuring — that BSEE FROM MIT your tester makes INIS6IANDIS Sense never goes STAFFSCIENTIST away completely. AT NATIONAL, 1 countered SEMICONDUC. with the story of TORCOR, the reference cir- SANTACLARA, cuit that had a CALE yield problem ‘The marketing manager wanted to take a big eon- tract for 1% trimmed parts, but the vield wasn’t very good for parts with a 1% tolerance. I got some data from the product engineer and studied it That's strange—the yield for 1% JANUARY 19,1991, parts was okay before trimming the output, ‘The problem tumed out to be a faise algorithm. (What is an algo- rithm? Isn’t it just a logarithm that ‘got twisted around? When I was a kid, we had lots of logarithms, and I think it's kind of suspicious that you don't see them any more...) And when we went baci and corrected the trim scheme, the yield was, of course, quite adequate. guess the moral of the story is that there are plenty of things that we shouldn’t trust blindly. I tell the test engineers that I'm going to check to see that things make sense when we're all done. And they had darned well better eheck on me, be- ‘ease nobody is immune from serew- ups, Certainly not me. Specifically, when you create a test, you have to make sure that the results make sense. If the answer is zero, and that doesn’t make any sense, then you should probably seta minimum limit. What if a part nor- mally has a power drain of 2mA, and 3,0mA maximum? What if yousawa Dateh of parts with less than 1/2-mA. current drain? Wouldn't you tend to get suspicious? What if we assumed the slew rate could be guaranteed by correlation with the power-supply current but there was no minimum limit on the test for current drain? A sanity check is in order! Some other time, we can think about other interesting cases that deal with ‘minimum lim its.” All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS: Mail Stop C2500A ‘National Semiconductor PO. Box 58090 Santa Clara, CA 95052-8090 DESIGNPEASE PORRIDGE WHAT'S ALL THIS RZAMBLE STUFF, ANYHOW? ometimes I go on hikes with my sons and my wife. Some- times T go on hikes without them; sometimes. they go on hikes without me. Here in Northern California, just in the San Franciseo Bay area, there are many dozens of parks, and many hundreds of miles of trails in these parks. Sometimes my wife leads a hike for the local chapter of the Sierra Club. Sometimes my son Benjamin (age 25, he has a full Red Gross cer- tifieate and I don't, so | am not eligi ble to lead hikes) leads the same hike. Thave been on some of their hikes. What they amount to is a Ramble. Or, as I mistyped the other day, a Rzamble, which is a word that sounds pretty good to me ‘The main point , that when you ‘20 on a hike with my wife, you will probably have a pleasant ramble If you go on the same hike led by my son, you will « have a pleasant "ramble, but it wi be 2 different BoBPEASE ramble. Ifyou join OBTAINEDA me on these hikes BSEEFROMMIT through the world IN1961 ANDIS of analog ideas, STATFSCIENTISY that will be a dif ATNATIONAL ferent Ramble, SEMICONDUC. ‘and I hope you TOR CORP. find it enjoyable. | SANTACLARA, read several daily CALE, columns in the San Francisco Chronicle—eolumns by Herb Caen, who has been writing daily columns ‘more than 50 years. Idon’tknow how he can do that! [also read columns by Jonathan Carroll, who is a little bit ‘meshugineh (alittle erazy}and quite EL amusing. Art Hoppe is another. I have big thick envelopes full of eol- umns by Stanton Delaplane and Charles MeCabe who wrote for the Chronicle for many years, but they died a few years ago. T must say, | have always been impressed with people who can assemble a few hun- dred words, everyday (or, every weekday).’This column I'm writing is every Ith day, and ¥ must admit, that ratio of LEI or 10:1 is a huge dif. ference. I maintain an awesome re- speet for people who ean put out a ‘column every day. ‘When I tookon this project, [knew immediately that if I had to meet a deadline, I would be in deep trouble. So, I would have to write a whole bunch of columns, and get way ahead of the game. Fortunately, I have had a little help from my friends, and T think I'm ahead of schedule—thanks to Frank Good enough, ELECTRONICDESIGN’s editor jn Boston. I have been encouraged and am making good progress at keeping ahead. I'm sure if start to fall behind, Frank will chew on my ankles and get me straightened out. How can I crank out all these words? Well, ithelps to havea decent word-processing machine to write on. Why am I typing this on an IBM- compatible Personal Computer (made by Compaq)? Well, T own an old Coleco ADAM word-processor at home and it works perfectly ade- quately for writing, memos and let- ters. But, it's not set up to transmit text encoded in ASCII, neither by modem nor by floppy, to Frank Goodenough in Boston oto the ed tors back in New Jersey. So I have this IBM-compatible machine, with processing, by PC-Write-Lite” from Quicksoft, which works pretty well. ‘And while I'm a great fan of ana: log, computers, T must say they're ECTRONIC DESIG JANUARY, 3881 not terribly successful at saving and storing and transmitting text. once set up a pair of voltage-to-frequency converters to put the X-Y coordi- nates of some letters and words onto fa stereo cassette recorder, It did work. I was able to store the words. But the resolution was marginal, the throughput rate was awful, and the amount of tape to store 100 words would be absurd ‘And when we played it back, using brace of frequency-to-voltage con- verters and a pen-plotter, the words and letters were shaky due to the jit ter and wobble and wow of the tine. base of even the best (analog) audio tape recorder. So, I'm not going to even try to use an analog-computer word processor—even though it’s not absolutely impossible. I'll use one of these new-fangled digital word-processors (which is not yet as user-friendly as my old ADAM) and plunk down my words. Itworks, and don’t gripe much about things that work. ‘What I really want is a word pro- cessor like that new Super Food Pro- cessor: You ean feed in a2 x 4, and the processor will grind it up into sawdust. Then you ean put the saw- dust into its hopper, and it will ex: trude them out into rigid 2 x 4, 1 just want to be able to do that, with words, too! All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS: Mail Stop C2500, National Semiconductor P.O. Box 58090 Santa Clara, CA 95052-8090 *PC-Write-Lite, available for $79 from Quicksoft Inc.,219 First Ave. N #224, Seattle, WA 98109—a very reasonable price, and a plausible, darned-nearly user-friendly piece of software. I recommend. 723. —Herb Caen just announced in June that—after 52 years in the game—he's going to cut back from 6 columns to 5 columns per week. That's still a huge number of words per week. But, no more “Sunday columns” from Herb Caen. nmPEASE PORRIDGE WHAT'S ALL THIS TEFLON STUFF, ANYHOW? nce upon a time, a long time ago, a friend of mine, Arnie Liberman, designed a really ‘good operational amplifier with a very low bias eurrent—less than 0.1 pA. Now, when you want to measure and (est a current as small as that, you don’t just measure the I X R drop across a resistor, because even with a 100,000-M0 resistor, it's hard to get resolution or accuracy, (0.1 picoamperes x 100 kM = just 10 mV).So Arnie set upa test with an integrator. If this amplifier had an output drift rate (or ramp rate) af 5 mV/switha feedback capacitor of 10 pF, that would prove that the ampl- fier’s input eurrent was 5 milli X 10 pico, or 0 femtoamperes But you ean't just run an integra- tor without reset- ting it occasional- ly. And Arnie knew you could not easily find a | relay that would short out the feed- back capacitor without introduc: ing lots of leakage and a bad jump when you turned EASE OBIAINEDA off thedrive to the BSEEFROMMIT — relay’s coil, So he IN3962 ANDIS__ made upa “relay” staresciennist by applying a le: ATNATIONAL ver to along push- SEMICONDUC- rod which reached TORCORP, _intothesealed box SANTAGLARA, where the test ‘CALE, was going on. The push-rod would close the contaets and short out tive capacitor. Then when you backed off on the push-rod, the integrator would integrate the current, and all you have to do is measure the out- put’s dV/dt to tell if the input cur- rent was within spec. And this worked fine, even at the level of ELECTR small femtoamperes. It worked fine in 1971...it worked okay in 1972...it worked in 1973, But after you started the test, you had to wait many seconds for the output Vat to stabilize to the point where you could get the right answer. Lat- er, in 1973, you had to wait almost a minute to get the right answer. In 1974, when youhad to wait 3or4 minutes to get any valid reading, somebody in Produetion Test finally complained to Arnie that you couldn't get very many units tested in a day. So Arnie went to trouble- shoot the test fixture. And he was puzzled for a while. What could cause such an erroneous reading? Soon, he realized that the push-rod ‘was the culprit. Ifyou pushed itback and forth, just a tiny bit, the integra- tor went berserk, Buthow could that be? The push-rod was made of Tef- Jon. And it slid in Teflon bushings. How could there be a big error caused by the push-rod? But as it turned out, that was in- deed exactly the problem. When the push-rod slid in its bushing, it gener- ated thousands and thousands of volts of static electricity. Now, on most ordinary insulators, those charges would drain off shortly. But because Teflon is such a good insula- tor, the charges would bleed off over a long, long time (through the air). Some of this charge would go into the summing pointof the amplifie-— for many seconds, As the Teflon got drier with time, the time constant got longer and longer until he had to fix it. Arnie re- placed the Teflon push-rod and Tef- Jon bushing with a grounded metal push-rod, in a grounded bush with just a tiny sliver of Teflon insu- lator on the end that pushed against the switch contacts. The problem was banished. It just goes to show that if you have the best materials, Oyre DESIG PEBRUARY 14,1991 and the finest concept, anid you mis- apply things justa little, you can get some terrible results. Tlecentiy, Twas testing some of my operational amplifiers, ‘LPC662's, with MOSFET inputs. The tine current wae consistently down jin the 2- or 3-fA area. It was so small that it really was hard to measure. I had a 10-pFintegrator running, simi- lar to Arnie's, but I was trying to get 20 times greater resolution. The res- olution was marginale there wore some jump nthe signal. he signal had Jong tails and additional errors if you tested a good amplifier after a Tad one had pegged. nd the sites ces didn't always give zero error on damp day. So I wrote down a list of the fixes J needed, and I set my new technician Paul to work on them. Paul made some imaginative and bold improvements. I had suggested that he should plot the ramp on a strip-chart recorder. Paul set up a digitizing scope and programmed it to spit out the answer sealed directly in omtoamnperes Tasked Pau) to set up a simple eir- caitto drive a 6 reed Flay so there wouldn't be much charge coupled Into te ireit rom the vltage that drives the coil, Paul set up a clever adaptive eireuit to take advantage of the small difference between the re- lay's pull-in and drop-out current, 80 that he fed the coil 2.6 V all of the time. You only had to hit the coil with a smait pulse of voltage, positive or negative, to close or open the con- tacts, respectively. I cautioned him that if the wires wiggle, they ean couple a lot of charge and noise into the input. Paul ‘strung the whole circuit up on rub- ber bands, as a shock mount, so that people walking on the floor nearby would not ruin the measurement. Also, I cautioned him that we would soon need a good, low-leakage Mockel, guarded wih Teton inst tion. Again, Paul gave me better than I asked for, with a big slab of Teflon for all of the components to be mounted on or above. T explained to Paul that the slow recovery from overload was caused by the silver-mica feedback eapaci- sitor which had poor dielectric absorp- tion. I gave him acapacitor that Thad zrade—filled with air—about 0.5-in.- by-2in-by-+in. of air, tomake a 5-pF cxpacitor. It had copper plates and guarded copper side-frames, ‘Now, first ofall, why am T telling, youalllof these details? If I design a tester with greatly improved perfor- mance to help me test a really bigh- performance produet, why should I tell all our competitors 0 that any- body in the world ean test thetr prod- ucts using the improved tester? Why should I give away all of these hard- earned secrets? Here's a preliminary answer: It's probably true that some competitors might learn how to test bet- ter and faster if I give them my techniques. But my eus- tomers will also learn fast- er. There are more of them, and it's more important to teach them, because they're still trying’ to come up the learning curve, There's not much point in keeping se- cerets about testing if our customers are kept in the dark. It’s just not fair to make your eustomers sruess what is a good way to test your products. Note that that is not the same as telling the eustom- er exactly how we testit (we have testers, big expensive ‘Teradyne setups, with monstrous in- terface boxes and complex software, and it would be much too complicated to tell everybody exactly how we run each test). But we do feel obligated to give any customer a good, valid test circuit that gives the same data as our production tests, ‘Okay. Paul set up this fixture. But there were a few litle problems. For one, the output ramp would oeeasion- ally give a jump at random times. This jumping seemed worse than previously. Also, lasked Paul to test a whole group of parts, and he said, well, it would take longer than one day to test those parts. OH?? [knew Paul could test those parts in less than a day—unless there was something wrong. Lasked, ‘well, why can't you test a part in @ ELECTRONIC PEASE PORRIDGE minute or two? He replied, “The out- put is offsseale and doesn’t, even come on-seale for 2 minutes.” That didn’t seem right, It wasn't until 3 days later that I slapped myself on the forehead at 6 A.M—it was the Teflon factor. Paul had added lots of Teflon around the cireuit, and the charge stored on its surface was the ‘major eause of the long tails. When T got to work, Tasked Paul to cover up as much of the Teflon as he could with aluminum foil—and ground it. He said the settling got a lot better. So he then removed all of the Teflon, and the slow settling went away, Just like Arnie's problem and Ar nie's solution, 16 years earlier! ‘Phe next problem was with those darn jumps. Now, we eould program the tester, not just to test the ramp for 60 seconds, but to test for six 10- second segments. Ifthe answer was the same for 3 or 4of the segments, then that is probably valid data and we should just ignore segments where the data had a big JUMB. Yes, we could do that. But, where did these jump errors come in? ‘The jumps were only 10 mV or so, always in the same direction, and at random times. Older fixtures did not have as many jumps. Why? The an- swer seemed to be related to that nice big feedback capacitor 1 had made. The cireuit did have fast set- tling—not such a long tail as the old- er silver-mica feedback caps. But Paul spotted some literature—some pesig "FEBRUARY 16801 Keithley data sheets—and some notes in Jiri Dostal’s book* about cosmic rays and charge. Dostal observed that when a cos- mie ray or alpha particle or other en- exgetic subatomic particle passes through matter, itoften eauses a charge of electrons. If there's a sen- sitive detector nearby, some charge ‘may come through the air and cause several thousand electrons to arrive atthe detector. Not just one electron, but several thousand—hey, that was the size of the jumps we were seeing—5 pf x 10 mV is about 50 femtocoulombs, equal to 300,000 electrons. So we re- searched a little more and realized that if we made a smaller volume of air adjacent to the delicate summing point of the amplifier under test, we should be able to cut down on the rep rate of these little jumps. So we're now prepar- ing two feedback capaci- tors. We're not sure which ‘one we will use, but they should both work quite wel One has an air dielectric, but it's only 1 in, by 1.2 in. with 0.080in. spacing be- tween the plates. It wilt have less than i/T0th of the volume of the old air eapaci- tor. The other feedback ca- pacitor will use about 5 inch- es of twisted pair, using one piece of bare bus wire, and another wire with sleeving made of...Teflon, Ifyou patit in the right place, Teflon is really good stuff. Some time I will tell more about which capacitor we ‘used, and the other details we need to test for femtoamperes. Sockets?? Relays”? Hall All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS: Mail Stop C2500 National Semiconductor P.O. Box 58090 Santa Clara, Ca 95052-8090 “Jiri Dostal, Operational Amplifi- ers, Elsevier Scientific, 655 Avenue of the Americas, New York, NY 10010; (212) 989-5800. About $113. sooPEASE PORRIDGE WHAT'S ALL THis CosT- ACCOUNTING STUFF, ANYHOW? nee upon a time, I was down at Lewis’ Restaurantin Nor- ‘wood, Mass,, waiting in line to order a’ Lewisburger, which was a fantastic hamburger with blue-cheese dressing and potato salad and ham. A voice eame boom- ing across the noisy restaurant, "Hi Bob.” I replied, “Hello Maurice, what's new?” Maurice wandered over and said, “How come you are still usimg the wrong formula, the old formula, for Factory Cost?” Old formula’ I never heard of a new for- mula, You mean the formula of Parts Cost + 450% of Labor Cost is obsoleted? Maurice said, “Oh, that’s been obsolete for months. “Didn’t anybody tell you? The new Factory Cost is 2 X Parts Cost + 3 x Labor Cost.” Well, T or- dered my Lewis- burger, and ate Tunch and ban- tered with my friends. But on the way back to my office, the wheels began go ing around in my \ | head. How could BOBPEASE they change the OWTAINEDA formula for Fae: BSEEFROMMIT tory Cost and not IN1961ANDIS__ inform peopie? starrscieNnisr How could they ATNATIONAL accept that, after SEMIGONDUC. —“‘a few months,” TORCORP. people were still SANTACLARA, using the incor- cau. reet formula, but it was no big deal? And they wouldn't bother to tell a guy unless they happened to bump into him ata restaurant? So I went to ELEcTRONTC see my boss, Richard Randlett, to ask him if he knew what was going on. He said, yes, he had heard about the new formula, but he didn’t know it was officially in effect. And, be- sides, he didn't want to bother us with unimportant details. We should keep on engineering our designs us- ing good judgment, as usual, and not worry about this “New Factory Cost Formula,” and other academic things like that. Nota big deal. Well, okay, I went back to work. But with: in an hour, I was back at Richard's office “Hey, Richard, you know that low- cost. converter Tam working on?” Yeah. “Well, I was going to use double-sided pe board, but with the new formula in effeet, it would be cheaper to use a single-sided board and a few jumpers. The parts cost will godown quite bit, but the labor will only inerease alittle.” Richard sat, there. He thought about it for a few seconds. Then he saw red! “No, don’t make that. change!!” I replied, “But Richard, the new Factory Cost Formula says it's wise toadda few seconds of labor tosavea loton parts cost.” But Rich- ard pointed out that even though that was what the bean counters were telling us todo tosavemoney, it ‘would be wrong. He told me to keep on optimizing the costs, assuming that Labor Costs would be multiplied by 4.5 compared to the Parts Cost. Meanwhile, he would try to get this resolved, and get the formula changed back to a reasonable pro- portion. He'd get mean answer with- ina month. Well, one month later, Richard DESIG FEDRUARY 28,1991 was still trying to get the problem resolved. Two months Tater he was gone, and three months later, I re- signed. So to this day, T don’t know how this was ever worked ovt. But 1 did know that could not do my job if nobody would tell me the rules, andif the rules kept changing to contradict, themselves. One set of rules told me that it was wise to add jumpers in- stead of plated through holes. And a second set of rules told me that was foolish and costly, and I should do it the other way. And if no one could tell me which rules were valid—hey, that’s a problem. I needed to solve that problem, but nobody eould help ‘me resolve it. So I voted with my feet and left that company, and I haven't worried about that kind of problem. Until recently. Of late, I was collaborating on a small regulator chip where we want- ed to trim the output to high aecura- cy. That would take about 10 Zener- zap trims, And those would tend to take up lots of die area—probably a ‘50 increase over the cir Conversely, if we put in trims, that would keep the die size small, but thecost for trimming each die would rise considerably. How much would it rise? We did not have the cost data. And when we got the cost data, it was kind of outdated, and we did not entirely b But the time for trimming with 10 - ser euts would basically take an ex- trasecond, due to the need to get the laser aligned with the die. And the costof testing ona tester witha laser is about double—about 4 eents/sec- ‘ond—versus 2 cents/second on ates- ter without a laser. ‘That is true, even for the tests ‘when you're not using the laser. This is because the whole machine must be paid for, even if you're not utiliz- ing all of the tester. So, 3seconds 4 cents/second is noticeably different than 2seconds x 2cents/second. For an Scent difference, you ean pay for amuch bigger die area! So that con- vvineed us there was probably a east advantage if we could avoid laser ‘trimming. But those considerations were ‘overruled by a more important con- Gl‘© Input Voltage 90 to 130 VAC (47/440Hz) © Single, Dual, Triple Outputs © 1200V Rms Isolation © Low Isolation Capacity Available © Continuous Short Circuit Protection © High Efficiency © Fully Regulated Voltage Outputs © Operating Temperature —25°C. to +70°C. with No Heat Sink or Electrical Derating Required * Expanded Operating Temperature Available (—55°C. to +85°C. ambient) © Optional Environmental Screening Available PIC Electronics, Inc. {EN Maecen Poy. ML var, RY 10552 Cal Tol Free 800-431-1064 vvew york ca, 914-699-5594 CIRCLE PEASE PORRIDGE sideration: The need to fit the die ina small package. There were some small plastic packages that the larg- er (cheaper) die would not. fit into. Still, we realized that in the future, ‘we will be sharpening up our pencils more often. Whatiis the right way to keep the costs low? What are the rrules in our business? We're not sure, but we're trying to get somre of these rules nailed down. Then when itis important, we will have the rules available—and believable. And that’s not.a trivial statement. What are the rules in your busi- ness? Do you know what they are? Do you believe them? Are they ever subjected to thoughtful serutiny, or are they so old and dusty that you know darned well that nobody has thought about them for many years? Do people change them without un- derstanding all of the ramifications and repercussions? Maybe it’s time fora sanity check. All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS: ‘Mail Stop C2500A, National Semiconductor PO. Box 58090 Santa Clara, CA 95052-8090 LETTER T0 BOB Dear Bob: The most disturbing thing to me about Spice is that many compa- ‘nies have adopted a philosophy of going directly from Spice to board design. You know, and fknow, that analog design has too many subtle traps—breadboards must also be built. Toe seen too many simulations where the problem wasn't lack of convergence, getting error mes- ‘sages, weird results, or weird wave- forms; but rather something even ‘more disturbing: Getting good simulation results on circuits that Tknew, or that breadboarding lat- er showed, did not and could not work worth a darn. JERRY STEELE, ‘Applications Engineer ‘Apex Microtechneiogy Corp. Tucson, AZ LECTRONIC¢ FEBRUARY 2, 1901, American dance is more popular chan evr and on ofthe Feb The New York Pubhc [brary Dance Collection. Choreogeapher Eliot Feld says the Libraat Lincoln Cee ie a vial a work a Iny studi Ages de Mile sys “the revival of any work is de- Bray bane Coleatont "And theyfe not the only ces Far dancers ae chore raph vr one 370 tolumes, 290,000 photog dix! an normous in archive ive bee essential elements in the renaissance of American dnc ‘Thurs just_one way. The Now York Pui ibrarys re sec seve The bey oes pays and’ puppet shows for chen, progetto the cf anne foreign angoage and ethnic lets, at scone journals ‘Again an! ane the, brary ccs out fe ra ‘The New York Public Library ‘AN OPEN BOOK DESIGNPEASE PORRIDGE WHAT'S ALL THIS STATISTICAL STUFF, ANYHOW? "ve always been a fan of Mark Twain and his writing. He had a rather good perception of the American people, and many top- ies that he wrote about are fascinat- ing to this day. One of my favorite quotes of Twain is: “There are three kinds of lies: there are “lies,” there are damned lies, and there are STA TISTICS....” ‘One thing that doesn’t help me a darned bit is “statisties,” at least in the sense that most mathematicians and engineers use them. I find most statistical analyses worse than use- less. But I do like to use charts and graphs. I took some data of diodes’ Vpversus Ip recently. The data wasa litle suspicious when I wrote down the numbers, but after [plotted the data, [knew there was’ something wrong. Then just went back and took more data |) until Funderstood what the error f was, ac current s noise that was be- . ing pumped outof | the inputs of the BOBPEASE digital voltmeter, OBTAINEDA crashing into the BSEEFROMMIT — diode,andcausing INISLANDIS rectification. If STAFFSCIENTIST data arises from a ATNATIONAL well-behaved phe- SEMICONDUC. nomenon and con- TORCORP, forms to a nice SANTACLARA, Gaussian distri- CALE, bution, then I don't care if peo- ple use their statistical analyses—it may not doa lot of harm. (Personally T think it does harm, because when ELECTRONIC you use the computer and rely on it like a eruteh, you get used to believ- ing it and trusting it without think: ing...) However, when the data gets serewy, classical statistical analysis is worse than useless For example, one time a test engi- eer came to me with a big formal report. Of course, itdidn't help thatit arrived at 1:04 P.M. for a Produetion Release Meeting that was supposed to start at 1:00 PM. But this was not just any hand-serawled report. It was handsome, neat, and computer- ined; it looked professional and com- pelling. The test engineer quoted many statistical items to show that his test system and statistical soft- ware were great (even if the ICs weren't). Finally he turned to the last page and explained that, according to the statistics, the ICs’ outputs were completely incompetent and way out of spec. Thus, the part could not be released. In fact, he observed, the median output of the output was 9 V, which was pretty absurd for the logieal out- put of an LM1525.type switching regulator, whieh could only go to the Low levelof0.2V or the High level of 184 V, How could the outputs havea median level of 9 V? How do you get an RS flip-flop to hang up at an output level half-way between the rails? Unlikely... Then he pointed out some other statis: ties—the 8 sigma values of the out- putwere +80 V and-8 V, Now, that’s pretty bizarre for a circuit that only has a +20-V supply and ground, (and it isn’t running as a switehing regu- lator, it's justsitting there at de). The meeting broke up before I could find the facts and protest, so that product Desig MARCH 14, 101 ‘wasn't released on schedule. Ttturns out, of course, that the tes- ‘ter was running falsely. So while the ‘outputs were all supposed to be set to +184 V, they were actually in a ran- dom state. Half of the time the out puts might be at 184 V and half of the time at 0.2 V. If you feed this data intoa statistial program, it ight in deed tell you that a lotof the outputs would be at +9 V, and some of the outputs might be at -8 V, assuming that the data eame from a Gaussian distribution, But if you look at the data and think, i's obvious that the data came from a ridiculous situa- tion. Rather than ramming the data into a statistical format, the engi- neer should have checked his tester. Unfortunately, this engineer had so much confidence in his statistical program that he spent a whole week preparing the Beautiful Report. Did he inform the design engineer that there were some problems? No. Did he check his data, check the tester? No. He just kept his computer erank- ing along, because he knew the eom- puter analysis was the most impor- tant thing. We finally fixed the tester and got the product outa litte late, but obvi- ously I wasn'ta fan of that test engi- neer (nor his statisties) as long as he was at our company. And that's just one of a number of examples I trot out when anybody tries to use statis- ties that are inappropriate do like to use seatter plots in two dimensions to help me look for trends, and to look for “sports” that run against the trend. I don’t look at lots of data on good parts or good runs, but Istudy the heck out of bad parts and bad runs. And when I work with other test engineers who have computer programs that facilitate ‘these plots, [support and encourage those guys to use those programs, and to look at their data, and to think about those data, I’ support anything that facilitates thinking, A couple years ago, I was ap proached by an engineer who was trying to use one of our good voltage references with a typical character- istie of about 20 ppm per 1000 hours long-term stability at +125°C. He olPEASE PORRIDGE was using it around room tempera: ture, and was furious because he ex pected it to drift about 0.1 ppm per 1000hours at room temp, anditwasa lot worse than that. He asked why our reference was no good. I pointed out that amplifiers’ drifts and refer- ences’ drifts do not keep improving ron! cer Ee eI by a factor of 2 every time you cool them off another 11 degrees more. T'mnot gure who led him to believe that, but in general, modern elec. tronic components aren't greatly im- proved by cooling or the absence of heating. In fact, those of us who re- member the old vacuum-tube days i oS ocat nego tg ee: nliherz available as prt Pry GU om aL err ee} rae Gy Seca ‘caRcLE 12 Blevectronie MARCH 1961 remember that a good scope or volt meter usually worked better if you keptit running nice and warm all the time, because aif of the resistors and components stayed dry and never ‘got moist under humid conditions. I vwon’tsay that the electrolytic eapae- itors might not have liked being a lit- tle cooler. But the mindless effort to improve the reliability by keeping components as cool as possible has been overdone, 'm sure you can blame much of that foolishness on MIL-HBDK-217 and all its versions Insome businesses, you have to con- form to 217, no matter how silly itis, but jn the industrial and instrument business, we don't really have to fol low its every silly quirk and whien. One guy who argues strenuously about -217 is Charles Leonard of Boeing, and you may well enjoy his writing (Leonard, Charles, ‘Is reli- ability prediction methodology for the birds?,” PowerConversion and Intelligent Motion,” November 1988, p. 4) Soif something is drifting alittle and you think you can make a digimprovement by adding fanand knocking its temperature down from +7 to +55, I caution you that you'll probably be disappointed be- ‘cause there usually isn’t a lot of im provernent to be had. It's conceivable that if you have a bad thermal pat tern causing lots of gradients and convection, you ean eut down that kind of thermal problem. In general, though, there's not much to be gained unless parts are getting up near theit maximum rated tempera: ture oF above 100°C. Even plastie parts ean be pretty reliable at +100°C. I know the ones I'm familiar with are his column is an exeerpt from the soon-to-be-published book T have written entitled “Troubleshooting Analog Circuits.” This endeavor will be published by Butterworths in Aprit 1991.) All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS, ‘Mail Stop C25008 National Semiconductor P.O. Box 58090 Santa Clara, CA 95052-8090 DESIGNPEASE PORRIDGE BOB'S MAILBOX Dear Bob: Sure enjoyed your Spice article in Electronic Design. I've followed ‘your musings for years, gosh, de- ‘eades. Please don'tever stop. There is another aspect of the simulation thing. Itis a matter of philosophy, T guess, and ithas concerned me for a Jong time: The sense of aecomplish- ment that comes from doing some- thing the first time. Onee you've done it, you've done it (how many endure the agony of climbing Mount Everest a SECOND time?) In the real world, this eonceptis intuitive, 1 think, Butin the world of simula- tion, we confuse what we have ac- complished with what we have imagined; worse yet, what the eom- puter has imagined! Dreams and re- ality mingle, and we are led astray. Case in point: Stirling engine de- velopment (one of my loves) has gone high tech. The talent and mon- ey is flowing toward the computer, not the lathe. We simulate, learn, simulate, ete. Each time we make an improvement its rewarding, and provides the psychological boost needed to continue. Because it isn’t sorewarding to do the same thing again, we don't bother to repeat in metal what we've already “built” on the sereen, Soon the simulation er rors compound until there is no way to distinguish dreams from reality sometimes wonder if we could have experienced the industrial rev- olution atalf, if somehow the eom- puter had already existed. Itis very tempting to simulate a thing rather than constructing it. ‘Simulation ain't the same as real- ly doing it, Perhaps a close parallel? Or, to say it another way, is it possi- ble that the boob tube in the office takes us on the same sort of jour- neys as the boob tube at home? DARRYL PHILLIPS President The Airsport Corp. Sallisaw, Okla. Teouldn't agree more. RAP BLEeTRONIC Dear Bob: just finished reading your column “What's All This Spicey Stuff Any- how? (Part 1)” in the Nov. 22 issue and I thought I'd share my own experi- ences with circuit simulation and EE education with you, Tam 38 years old, and was trained in math and conapater science. Two ‘years ago, [decided to pursue a Masters degree in EE ata well-known northeastern university. I told the boss that I was going to specialize in the mathy end of BE, like Control Theory, Signal Processing ete,, but Thad a hidden agenda. I wanted to learn electronics. It's all worked out because I have learned electronies, but only sort of. You see, at this institution's outlying campuses, where they erank out BSEEs and MSWEs by the score, there are no labs. I repeat, no labs. No breadboards, no seopes, no components, nonothing. Any design that takes place is on the computer using a cireuit simulation program. Ihave suggest- ed to several professors that a real live lab would be very helpful, but they just “hurrumph” about the eomputer being just as good as a learning expe- rience, So you see, it's not just lazy (young or otherwise) engineers eopping, ‘out on the design process, it's the edueational system taking the path of least resistance. Thave improved the situation for myself somewhat. I designed and built a de power supply (you're probably saying “big dea,” but that’s pretty good for & mathematician!) I bought a breadboard, a selection of components, and a digital meter. With this setup, I ean do quite abit of experimenting on my ‘own. My father, an “old tyme” electrical engineer, isso glad that his math- ‘ematician son has seen the light that he's salvaged an old seope from who knows where for me. I’m looking forward toits arrival. ‘As faras my experiences with the circuit simulation program are con- cerned, Iconeur with your opinion. We didn’t use Spiee, but some other Spiee-like program. It was graphically oriented and very user-friendly, but ithad the same crazy eonvergence problems that you describe. One project we had was to design an analog multiplier. Changing the quiescent current through the Gilbert cell ever so slightly made the difference between fast convergence for steady-state analysis and sitting for 15 minutes watehing the cursor bfink. Then, to add insultto injury, the ease that converged for steady-state analysis didn’t converge for Fourier analysis. I finally was forced to give it up. Interestingly enough, my professor accepted “Program Did Not Converge” as a correct answer, CHRISTOPHER LENNON Bedford, Mass. Ouch! This guy's heading in the right direction, but the schools arent, -RAP Dear Bob: feel that your most recent column (Spice, Part if} needs some comment. You say, “I don’t think you can beg or steal or borrow or buy a model of a transistor that’s guaranteed.” First, to what kind of transistor are you re- ferring: bipolar, JFET, MOSFET, small signal, large signal, power, micro- wave, discrete, integrated devices? Second, your statement implies thata single “transistor” model should be guaranteed to “ran” under all conditions. Suppose I suggest that you should be able to design a single OP AMP that works equally well for all ap- plications? I suspect that you would not only laugh in my face but that you ‘would think to yourself that I am a fool for having made the suggestion in Desi NED MARCH, 191RCI's REAL TIME DISK FAMILY GIVES YOU A CHOICE 2 [DS 323) 684 is 24905 322) 513 ose] | 45.6 DS 163 (DS 321 Ma los i23)os 6zjos zai] | 56 [os 122| | ns sea] bstei] _(os320 W DS.63 | DS 82 [DS 121} 15 240) 128 0S 62 s§ 4 0543 sé) (DS 160} as ps42[os6r| _[osta9 57 [0523 |ps41 |_| D580 42 osm [ose 28 0521 /0540 14 [Ds20 i 44° 89 134 179 269 359 538 719 Sustained Transfer Rates (MBytos/sec) ‘opacities (GBytes) C's NEW DotoSTORE family of rea ne dks allows you to choose the sped ond opacity yeuroppicton. Al modes ofthe DtaSTORE foil, froth smallest OS 20 through he largest ond festest DS 323, lz he sme quclty modular components wich means ol models ore upward compatible and expondebl FEATURES Bh Low cost i Compod size 1 Dsishese management sofware Low power ‘WE Interfaces available to VMEbus ‘Obus, Unibus, PC / AT bus 1 Highly celble Uses industry stondrd and feld proven dks OPTIONS Removable, uggedied asters 1 ntegted Bm archival tape High sped digitzers 1 Naworsinrfaing COR 601 / SMPTE RP 125 interface Wedded processor ACs an indy inovotr in proiig el ime action vitalimage procesig, and bigh speed steroe systems =~ uy Recognition Concepts, ie. IMAGING WITH SPEED wise .0, Box 8510 Inne Vilage, NV 89450 1-800-2GET RCI Phone: (702) 831-0473 FAX: (702) 831-8035 caReLe 301 CALL OR FAX US FOR IMMEDIATE INFORMATION PEASE PORRIDGE the first place. Well, to suggest that asingle transistor model should “san” under all conditions is equally laughable. Am Ihitting below the belt? ‘Third, you seem to want toplace the total responsibility for the sue- cess of the model upon that person who developed the model. However, I get the impression that you have no sympathy (or respect?) for a user of one of your op amps who misuses your op amp. Below the belt again’? Fourth, yau seem to believe that those “transistor” models which have been built into the circuit sina lation program Spice represent the complete set of models from which ‘one can choose. The sad truth is that those “transistor” models built into Spice represent just about the worst setof models from which one can choose, Why is that the ease? Be- cause industry gets those models for free from the Univ. of Calif. at, Berkeley as the result of student, slave labor and typically is not will- ing to pay the price for truly physi caland accurate models ...You suggest that you may wish to BUY a guaranteed transistor model. Well, you can. My company has developed and markets Spite (SPL Transistor Emulator), the world’s best C1S/SIS (MOS if you prefer) transistor model. The Spite ‘model does employ physically mean- ingful parameters and parameter values. Italso simulates, with un- precedented accuracy, a transistor which is properly used. Would you ike to guarantee me that your op amp will always “ran” for me no matter what I do to it? There just isn’t any point in building a lotof un- necessary complications into the model to protect the unknowledge- able user from himself. So what do you want your guaran- tee to say? Send mea reasonable spevification and a purchase order aan@\S wil write your guarantee for you DR. JAMES B: SMITH President Semiconductor Physies Inc Escondido, Cait Tuse mostly bipolars, but for CMOS this looks tempting!-RAP MevectrRoNic pesiteN MARCH 26,198)PEASE PORRIDGE WHAT'S ALL THIS PERFECTION STUFF, ANYHOW? nce upon a time, a lady went shopping for shoes. She ‘went into a shoe store and said to the salesman, “My good man, I would like to buy some alligator’ shoes.” The salesman showed her some alligator shoes. But ¢his one had a flaw, and that one had a blemish, and that one had seratch. He went back and got more shoes. But when the lady inspected them, she found an imperfection that was unacceptable to her every time. Finally, he brought out every pair of shoes he had, and she was not pre: pared to buy shoes with such de- fects. In great exasperation, the salesman told her, “Look, lady, you're not perfect. I'm not perfect. How do you expeet an alligator to be perfect?” One of my late est ventures in- volves the writing of a book. On Oc- tober 30, 1990, T handed ina com- plete, polished manuscript. I don’t know how many words, or BOBPEASE — how many pages, OBTAINEDA but it was about BSEEFROMMIT 700,000 bytes, and INI961ANDIS__ the Fed Express SIAFFSCIENTIST guy had to strug- ATNATIONAL gle to get it on its SEMICONDUC — way. This was a TORCORP,, refinement of SANIACIARA, some stories on CALI the troubleshoot ing of analog ei cuits I wrote a couple years ago. 1 took the old stories and added alotof new info, tweaked and refined a lot EL ECTRONIC of the text, and corrected some typo errors, finally rounding it into pretty good shape. With a little Iuek, the book will eome out on schedule this month. On November 14, Isent to the pub- lisher a package of new corrections and changes. Only one envelope, but there were 23 items in there. When I talked with the publisher, I could tell she was smiling, seeing that I was ‘willing to putso much effort into get- ting things right. On November 80,1 assembled an envelope with more corrections, changes, refinements, and upgrades. I mean, some of the people had moved in the last month, so certainly had to putin an address correction. In several eases, the dis- tributors who sold things didn’t sell them any more. In other instances, I remembered a new technique that I had forgotten previously. So, when I send in this package, I'm really mak- ing progress—because there are only 27 items. Next week I expect toremembera couple more items. I’m not sure ex- actly when the smile will completely disappear from my publisher's face, butit’s only amatter of time. She will say, “It’s time to shoot the engineer, and put this to bed and print what ‘we've got.” And I'm sure I will have to agree with her pretty quickly. But meanwhile, I know I have to correct the spelling of one guy's name. We can't leave that wrong. And an old friend is mailing me some more info on some diodes...and there are curves that don't look quite right and there are some photographs that still have to be developed, not to mention the ones that haven’t yet been taken... APRIL, 1091 “Time to shoot the engineer.” "That's a phrase that has been around for «long time, Almost as soon as I «got out of school and into industry, I ‘began to hear people explaining that the need for perfection was all very fine, but it must not go on much fur- ther, When is the ciresit going to be good enough? Perfection isn't neces- sarily justifiable. What is good ‘enough? And whose opinion is tobe relied on? Sometimes the engineer is correct that there are some improve. ments that have to be made, Other times, it’s not so clear. For example, we were recently teying to release a new produet, but the distribution of one parameter was not quite centered. The yields might occasionally fall off more than we liked. So we proposed an optional metal mask that would bring the dis- tribution back elose to center, But this might eause some problems at high temperatures. And it might cause some dynamic problems. And even if it didn’t cause serious prob- Jems, it might get some of our cus- ‘tomers cross if they had tore-qualify our product, because we had made a change in & mask, even though it would bea very tiny change. ‘What's the right thing to do? Ac- cept the yield loss? Change the data sheet? Delay the release of the prod- uct and risk the loss of market share? If we wanted to compromise, where would we do it? Hey, I don't know how you run your business, but in our basiness, there aren't any easy questions. If you're making all gator shoes, and your QC depart- ment says you can ship only shoes that have no visible flaws under a magnifieation of 6X, you've not go- ing to be selling @ lot of shoes. But, most of the time, our shoes are on our feet, at least 8 or 4 feet away from our eyes or anybody else's eyes. From that vantage point, “im- perfection” is quite different. ‘Now, I’mnot proposing that were- fuse to ship amplifiers beeause they're not “perfect.” I mean, if a perfect amplifier is one that has less than 0.5 mV of offset voltage, we have a lot of perfect amplifiers. Just 10years ago, customers were pretty pest 6 NEDPEASE PORRIDGE happy to buy that kind of “perfect amplifier.” But these days, even 15 pV isn’t “perfect.” And if you build lots of amplifiers with less than 10 HV, and they all test out good, and then you allow them to warm up, and you temperature eyele them through an oven, are they still “perfect” — better than 15 wV? Tf you want to buy a “perfect” amplifier, do you re- quire big safety factors against ev. ery possible condition? You may ‘wind up going barefoot:The price for a “perfect” pair of shoes might be more than you would be willing to pay. Sometimes we have considered that amplifiers at this level of preci- sion might have a looser AQL than normal—perhaps 0.5% instead of the ‘typical 0.01%. But our QC people don’t want to concede that. If we test an amplifier for noise, and we do it 30 times, the data might show low noise on 27 of the passes. But on 2 or3, there might be a small deviation—an inerease (or a de- crease) in the test result—whether or not the amplifier actually made more noise. Heck, you can’t design a noise test. to be perfectly rep able—that’s inherent in the nature of noise. Now, if a unit passes a test at its “class’” or final test, it must never be allowed to fail its guaran- teed specs if you re-test it. That ‘means you must have wide guard- bands, as wide as the deviations of the system's noise. If an amplifier that reads 1 pV pk-pk most of the time ean read 0.8 nV some times, and 12 j V other times, that would tell you that you must havea guardband Of at least 0.4 uV, and maybe 2 or 3 times that, So you eould sell to a guaranteed spec of 2.2 pV—even though most of thepartsare1 .V pk- pk. 1s thatithe spee you want to buy? Onee, a long time ago, a bright young, engineer was working on semiconductors in England. The transistors were passing @ 27-V breakdown test very reliably. But this semiconductor company was a subsidiary of a large plumbing-sup- plies company. And if you're making boilers or gauges or pipes, you've got to have a 3:1 ratio between the working pressure and the bursting pressure. So, this semiconductor ‘company wrote their data sheets for 9-V transistor, while their competi- tors were selling to a 25-V spec Needless to say, the young engineer knew there was no future in a busi- ness where perfection and safety factors make the playing-field so badly tilted. And that was why Bill Frusztajer left England and eame to the United States, where he became head of Teledyne Crystalonics All for now. / Comments invited! / RAP / Robert A. Pease / Engineer ADDRESS: ‘Mail Stop 025008 National Semiconduetor PO. Box 58090 Santa Clara, CA 95052-8090 Oa rece un) ee Aas Ceara Biel uence inure ari i Peer Pei eas Sno Pease Cane culied PCa en at Prue te ly ace fog eens Offered in industrial, eee) re eRe cae Pee ernest neers sci Soe a a WET ele as ‘CIRCLES? Peery Deron) Me vectrowic Cee ernie) "APRIL 101 AUTOCAD for Electronic Engineers Only $195 Naw Symool icon browsing Unlimited levels of hierarchy Spice & Susie interfaces ‘AutoSchema’ AutoPCB Best performance on a P.C. Double sided SMT Real time design rule check * Interactive push & shove routing AutoHybrid ‘+ Worlds only P.C. Hybrid system ‘Automatic component synthesis Custom die geometry + 05 micron resolution ‘CADISYS: aus cnmey Pace, cat roR Sey Soe 00 PAX os) 431.0900 EXT 200, cRCLE Ls DESIGNPEASE PORRIDGE Waat’s Aut Tuis MENTORING STUFF, AnyHow? nice upon a time, a new en- igineer came to work in our group. A woman. Now, in Some areas, it’s really not a surprise to have a new engineer o a womanengineer, butinourgroup, that did not happen very often. So when Jane arrived, we all tried to be polite ‘and cheerful, forachange,andnot just sereamatherand give her ahard time, sas newcomers are sometimes treated Now,Janewasabright young woman, but there were a lot of things that she had to ask questions about, so she ‘would ask various people, Sometimes she would ask me, and sometimes sive would ask Andy, another of the old time experienced engineers. We