Asic Design
Asic Design
Himanshu Patel
Space Applications Centre (ISRO) [email protected]
Contents
o o Introduction ASIC Design Methodologies n Full custom n Standard Cell n Gate Array ASIC n Structured ASIC ASIC Design Flow n Design Entry n Functional Verification n Synthesis n Design For Test (DFT) n Place & Route n Timing Verification n Formal Verification n Proto ASIC Test Mixed Signal ASIC Challenges for Deep Submicron ASIC CASE Study : OBC ASIC
2 Himanshu Patel
o o o
ASIC
o ASIC stands for Application Specific Integrated Circuits. o It means an integrated circuit designed for a specific application. o An application could be a microprocessor, cell phone, modem, router, etc. o Nowadays, ASIC has a complete system on it, often called as System on a Chip (SOC)
Himanshu Patel
Full-custom design
Each
Standard-cell based design Standard library cells (NAND,NOR, XOR,FF etc) are used for design
Gate-array based design Wafers are pre-fabricated with unconnected gate arrays
Himanshu Patel
o Pre-defined library cells (NAND,NOR,FF,RAM, Hard macro cores etc) are used o Designs are created using schematic capture or synthesis from Hardware Description Languages (HDL) o All mask layers are customized transistors and interconnect
ASIC Design Flow 6 Himanshu Patel
Himanshu Patel
channeled gate array: The interconnect uses predefined spaces between rows of base cells
ASIC Design Flow 8
channelless gate array: Only some (the top few) mask layers are customized
Himanshu Patel
Himanshu Patel
Structured ASIC
o A Structured ASIC falls between an Gate Array and a Standard Cellbased ASIC o The design task involves mapping the design into a library of building block cells, and interconnecting them as necessary. o Largely Prefabricated n Components are almost connected in a variety of predefined configurations n Only a few metal layers are needed for fabrication n Drastically reduces turnaround time
10
Himanshu Patel
Structured ASIC
Advantages: o Low NRE cost o High performance o Low power consumption o Less Complex
n Fewer layers to fabricate
Disadvantages: o Lack of adequate design tools o Design constrained by pre-fabricated block available in library
ASIC Design Flow 11 Himanshu Patel
Comparison Graph
Volume
12
Himanshu Patel
3. ASIC Fabrication:
o GDS-II to ASIC chip o Done by foundry
13
Himanshu Patel
IP Cores
Design Entry (RTL Coding) No BIST & JTAG Insertion (DFT) Synthesis
Testbench coding
Yes
DRC Error
14
Specifications
o The chip functionality is described in Requirement & Specification Document o The targeted speed, power consumption, area are also specified o System Engineer conveys requirement in plain English to Design team and Verification team
n Design Team generates RTL code as per specs. n Verification team generates Test benches/test cases as per specs
Back
15
Himanshu Patel
Design Entry
o Either by Scematic Capture or through HDLs like VHDL,Verilog etc o The quality of final chip depends largely on quality of RTL code o There are some design guidelines which should be followed
n n n n Design should be synchronous Clock gating should be avoided Flip flops should be used instead of latches Proper FSM coding styles (one hot, binary, etc)
o IP Cores or third party soft cores are used for standard blocks like processor, MAC, UART etc
ASIC Design Flow
Back
16
Himanshu Patel
Testbench
Unit Under Test
Stimulus Generator
Output Monitor
Test Report
o First Test plan is worked out based on which different test cases are identified
o Assertion based testbenchs checks captured output with expected output and writes report
ASIC Design Flow
Back
17
Himanshu Patel
Functional Verification
o The functionality of RTL code is verified using testbenchs it is also called behavioral Simulation o Some of the popular simulators : n ModelSim n NCSim o Code coverage indicates how much portion of RTL code is covered by testvectors n Statement coverage n Expression coverage n Branch Coverage n Toggle coverage o Typically a good testbench achieves more than 95% code coverage
Back
18
Himanshu Patel
o DFT methodology
o DFT Advantages:
n Improve quality by detecting defects n Make it easier to generate vectors n Reduce vector generation time n Area overhead of 10-15%
19 Himanshu Patel
o DFT Disadvantages:
ASIC Design Flow
MBIST
ASIC Design Flow 20 Himanshu Patel
JTAG Insertion
IEEE 1149.1 standard for Boundary Scan test
Back
21
Himanshu Patel
Synthesis
o Process of converting RTL code in to gate level netlist o ASIC Vendor provides Cell library of basic gates (AND,OR,FF,RAM,FIFO..) o Pre synthesized IP Core blocks (DesignWare,) are treated as Black Box o Some of the popular synthesizer :
n Synopsys DC, Cadence Ambit BuildGates Synplify ASIC
ASIC Design Flow 22 Himanshu Patel
o Special care should be taken for high fanout nets like clock & reset
n They are not synthesized at this level Set_dont_touch_network n During Clock Tree Synthesis process, layout tool creates optimized clock tree
ASIC Design Flow
Back
23
Himanshu Patel
o Delays are specified as (min, typ, max) depending on PVT (Process, Voltage, Temperature) condition
o Wireload models are used to estimate propagation delay based on fanouts because at this stage Layout is not done o Setup violations must be addressed
n Pipelining n Register retiming (balancing combi. Logic)
Setup violation: Data late, clock early Max delays are considered Decides maximum clock frequency Hold violation: Data early, clock late Min delays are considered
Back
25
Himanshu Patel
n Undriven nets n Naming convention errors o DRC tool kit is provided by ASIC foundry
Back
26
Himanshu Patel
Timing Constr.
Placement
Pre Layout N/L Post Layout N/L Formal Verification Equivalence check
Scan Chain Stitching & ATPG (DFT) Clock Tree Synthesis Routing Back Annotation SDF
(RC Delay Extraction) ASIC Design Flow
STA
Netist
27
GDS-II file
No
Timing OK ?
o scan cells are NOT connected until placement is completed so chain is not formed at this stage
28 Himanshu Patel
Floorplanning
o Floorplanning is a mapping between the logical description (Hierarchical Netlist) and the physical description (the floorplan). The goals of floorplanning are to: o arrange the blocks on a chip, o decide the location of the I/O pads, o decide the location and number of the power pads, o to minimize the chip area and delay
ASIC Design Flow Himanshu Patel
29
Floorplanning
initial random floorplan generated by a floorplanning tool
Back
30
Himanshu Patel
Placement
o Placement is arranging all the logic cells within the flexible blocks on a chip. o objectives of placement n Guarantee the router can complete the routing step n Minimize all the critical net delays n Make the chip as dense as possible
31
Himanshu Patel
block A contains four rows of standard cells (A.211, A.19, A.43, A.25)
Back
32
Himanshu Patel
Scan chain stitching & ATPG o After placement , Scan cells are stitched together to form a scan chain
o Normally Different scan chains are formed for different clock domain Flip Flops
33
Himanshu Patel
34
Himanshu Patel
Back
35
Himanshu Patel
Routing
o Routing is done in 2 steps n Global Routing : plans channels for routing between blocks, Its goal are: o Minimize the total interconnect length. o Maximize the probability that the detailed router can complete the routing. o Minimize the critical path delay.
Back
37
Himanshu Patel
38
Himanshu Patel
Formal Verification
o Equivalence check between pre-layout and post layout o Mathematical models are made to check functionality equivalence at each node of netlist o FV can also be done between RTL & Netlist o EDA Tool
n Formal Pro (Mentor) n Formality (synopsis)
Back
39
Himanshu Patel
o Compared with dynamic simulation, STA is much faster because it is not necessary to simulate the logical operation of the circuit.
40
Himanshu Patel
Timing Paths : o Input path (I/p pad to FF) o Sequential path (FF to FF) o Output path (FF to o/p) o Combination path (i/p to o)
ASIC Design Flow
Back
41
Himanshu Patel
Back
43
Himanshu Patel
44
Himanshu Patel
45
Himanshu Patel
46
Himanshu Patel
47
Himanshu Patel
49
Himanshu Patel
DW8051
Synopsys DesignWare 8051 Soft IP Core Features o Highspeed architecture :4 clocks per instruction cycle 2.5X improvement over the standard 8051 o Dual data pointers o 3 Timers, 2 UARTs o Extended Interrupts (7 nos) o Variable length MOVX to access fast/slow RAM peripherals o Fully static synchronous design
ASIC Design Flow 50 Himanshu Patel
51
Himanshu Patel
o Timing Signal Generator: This is generic timing signal generator, which can generate up to 16 programmable pulses o Auxiliary Data Interface: This is parallel/serial auxiliary interface with built in dual port RAM. o Serial Synchronous Transmitter/Receiver This is 3-wire (clock, strobe, data) synchronous tx/rx o Monitor Program :OBC ASIC contains 1K Bytes of on chip ROM which holds Monitor program firmware.
ASIC Design Flow 52 Himanshu Patel
o In Phased array distributed controller it is required to load Transmit & Receive characterization data within time constraint as shown above o This task was earlier implemented in software as Interrupt Service Routine, but due to variable interrupt latency it was not meeting timing constraint o So a Hardware module was implemented
ASIC Design Flow 53 Himanshu Patel
EEPPI - Architectute
clk reset_n
8 Port0_PPI [7:0] Port1_PPI [7:0] Port2_PPI [7:0] Port3_PPI [7:0] Port4_PPI [7:0] Port5_PPI [7:0]
8 8 8 8 8
Data out[2:0] Look Up Table Addr (LUT) (Event -Action) 6 64 X 3 bits Flags 2
MUX
54
Himanshu Patel
UART_RRS
UART with Recursive Running Sum Filter to remove noise samples from incoming serial data
55 Himanshu Patel
UART_RRS has better performance than standard UART at higher Noise levels
n n UART_RRS can decode data correctly up to 37% corrupted sample Standard UART can decode data up to 6 % only
56
Himanshu Patel
58
Himanshu Patel
Thank You