Chapter 4: Programmable Logic Devices

4.1

Chapter Overview
This Chapter provides an overview on Programmable Logic Devices (PLDs) form

the history of programmable logic devices to the device types. PLDs come in two forms,
Complex Programmable Logic Devices (CPLDs) and Field Programmable Gate Arrays
(FPGAs) both having their advantages and disadvantages with respect to the specific
application or design they are to be used in. The PLD used for our design, which was a
CPLD from Lattice Semiconductor is discussed. There is a need for design of smaller
more dense electronic designs requiring less board space as well as less power in the
space and military industries there are very few PLD vendors that provide Radiation
Hardened components.

Unfortunately intellectual Property Cores will be discussed

which is the latest trend for vendors to provide system-on-a-chip capability for their
FPGAs.
4.2

Introduction

4.2.1

Background of Programmable Logic Devices

A programmable Logic device refers to any type of integrated circuit that a logic

design can be implemented and reconfigured in the field by the end user. Since these
logic devices can be programmed in the field they are also called Field Programmable
Logic Devices (FPLDs). The PLD provides flexibility for designers to implement many
different designs in varying complexities for many different applications. One of the
most common PLDs is the one time Programmable Read-only Memory (PROM). This
comes in two different types: (a) mask programmable devices programmed by the vendor
using a custom mask and interconnects and (b) field programmable devices that are

16

configured by the user.

One of the great advantages of PLDs is that they are very

inexpensive at low quantities.

A device that was a follow on from the PROM technology that can be used for
logic designs was the Programmable Logic Array (PLA). The PLA using the PROM
structure turned out to be the first Field Programmable Logic Array (FPLA). The first
FPLA was introduced in the mid-1970s. The FPLA had a fixed number of inputs,
outputs and product terms that consisted of AND and OR arrays that contained
programmable inputs. The FPLA did not have great success because they were very slow
and complicated to use. The designer had to design to a fuse map instead of conventional
boolean equations or schematic capture.
In the late 1970s the Programmable Array Logic (PAL) architecture was
introduced that increased the use of programmable logic. The PAL architecture consisted
of a programmable AND array and a fixed OR array so that each output is the sum of a
specific set of product terms. The design entry tool for the earlier PAL was in the form
of Boolean equations making it very easy to learn and implement. PAL devices are now
available in different varieties from different vendors providing flexibility inputs/outputs,
size of the OR-gate, and flip-flops. Some PALs are even provided in either
NAND/NAND or NOR/NOR structure to increase design flexibility instead of the
AND/OR structure.
PLDs can be divided into two groups, Simple Programmable Logic Devices
(SPLDs) and High-Density Programmable Logic Devices (HDPLDs). SPLDs come in
the PAL and PLA architecture, while HDPLDs include CPLDs and FPGAs. Figure 4.1

17

contains a hierarchical block diagram of the PLD architectures, subfamilies and

programming technologies.
Programmable
Logic
Devices (PLDs)

Simple
Programmable
Logic
Devices (SPLDs)

High-Density
Programmable
Logic Devices
(HDPLDs)

Programmable
Logic Array
(PLA)

Programmable
Array Logic
(PAL)

Bipolar
Technology

CMOS
Technology

Complex
Programmable
Logic Devices
(CPLDs)

EEPROM-CMOS
Technology

Field
Programmable
Logic Devices
(FPGAs)

UVEPROM-CMOS
Technology

SRAM
Programming

Antifuse
Programming

Figure 4.1 PLD Hierarchical Architecture

4.2.2

Simple Programmable Logic Devices

The simple PAL architecture has become an industry standard. PLAs and PALs

that have pin packages of 20 44 pins and density ranging from 100 to several hundred
gates are considered SPLDs. The Basic AND/OR architecture PALs are flexible devices
that can implement logic equations in Boolean sum-of-product (SOP) form.

Some

enhancements to SPLDs have been programmable input/outputs (I/Os), bidirectional

I/Os, programmable output polarity, flexible register configurations and chip clocks.
18

One important advantage for PLDs is that they can replace small to mediumscale integrated (SSI/MSI) circuits for higher packaging density. One PLD could replace
tens of integrated circuits with 200 500 gate equivalent. Other benefits for SPLD is that
they reduce power, have faster turn-around time, faster performance because they reduce
interconnects between chips and higher reliability. SPLDs are available in bipolar and
Complementary Metal Oxide Semiconductor (CMOS) technology. In CMOS technology,
they come in Erasable PROM (EPROM) based which are Ultraviolet Erasable
(UVEPROM) and Electronically Erasable (EEPROM).
Due to the simple architecture of an SPLD they offer very high performance. The
SPLD devices are at 0.5 um CMOS process with logic delays down to 3.5 ns and
frequencies as high as 200 MHz. Higher density devices are coming on the market in the
area of Complex Programmable Logic Devices (CPLD) with high performance, but
SPLDs still have the best performance, easier to use and design with because they are an
industry standard.

Computer networking components and other telecommunication

equipment still demand the need for SPLD devices due their high performance. Because
of the trend to migrate to higher densities SPLDs have been driven to specialty markets
such as cellular phones, video games and hand held web browsers.

Table 4.1 shows a

comparison of SPLDs and CPLDs over a broad range of criteria.

New innovative advancements in SPLD include programmable output logic,
macrocells that can be configured as combinatorial or register operation and active low or
active high polarity. SPLDs are also available in lower-voltage and low power offering 5
volt and 3.3 volt devices. This allows more flexibility for design applications that require
trade-offs between low power, high frequency and low voltage.

19

Criteria
Propagation Delay
Density

SPLD
High Speed - Typically
3.5 ns
100 to Several Hundred

Technology
Ease of Designing
Complexity

Bipolar and CMOS

Very Easy
Simple Architecture

Frequency
Programmable I/Os

200 MHz
Yes

CPLD
High Speed - Typically
10 ns
Several Hundred to several
Thousand (25,000 Gates)
CMOS
Medium Ease
Medium to Difficult
Architecture
100 MHz
Yes

Table 4.1 SPLD and CPLD Comparison

4.2.3

High Density Programmable Logic Devices

The main disadvantage for the SPLD is an architectural limitation. SPLDs have a

limited amount of logic structures that can be allocated in a design in a fixed way. Highdensity or high-capacity PLDs (HDPLDs/HCPLDs), which are also called complex
PLDs (CPLDs) and Field Programmable Gate Arrays (FPGAs), try to solve the silicon
limitation by adding more flexible block structures and interconnects. Programmable
Logic Devices include simple as well as high-density PLDs.
In 1985 Xilinx Corporation came out with the first FPGA. It introduced the Logic
Cell Array (LCA) and was the building block for all FPGAs to follow. It contained a
pool of independent logic cells and multiple routing resources that allowed any logic cell
to be connected to any other logic cell. This provided routing flexibility. The first LCA
consisted of a combinatorial logic function and a flip-flop.

The I/Os could be

programmed as input, output and bi-directional and the routing resources allowed for
distributed clock through the chip, high speed and low-skew signals and local routing
from LCA to LCA.

20

The two major elements of CPLDs and FPGAs are the logic elements and the
interconnect structure. The logic elements are also known as macrocells, logic cells
and/or logic blocks. The interconnect structure is how those elements are connected
together to perform the design for a specific application. As mentioned there are two
high-density programmable logic devices.

Those are complex programmable logic

devices (CPLDs) and field programmable gate arrays (FPGAs). It is hard to determine
the difference between the two but usually in CPLDS there is fixed routing resources onchip and routing is done via a switching matrix, which leads to predictable delays. In the
SPLD architecture, each macrocell contains its own product term. However, in the
CPLD architecture the vendor takes advantage of the complex macrocells and employs
product term steering or product term sharing between the macrocells. The term complex
in CPLD refers to pin count and the amount of internal macrocells. The vendors try to
provide an output pin for each input set, which increases the complexity.
One of the main performance criteria for PLDs is Total propagation delay (Tpd),
which is the delay from the input to the output pin through a specific function. On
SPLDs of 44 pins this value is fixed and predictable due to less complex designs and is
averaged to be about 5 ns. When there is a larger and more complex design the Tpd can
be as high as 7.5 ns on devices that have 100 pins and 128 macrocells.
In the past, there has not been a clear distinction between the roles of CPLDs and
FPGAs until recently.

A CPLD used to be higher priced and as mentioned have

predictable timing specifications making them ideal for high-speed applications. FPGAs
where more reasonably priced and satisfied the requirements for low to medium end
performance applications. FPGAs have found a niche for space and military applications

21

due to the availability of radiation Hardened devices. This will be discussed later in the
chapter. Recently, CPLDs have been reduced in cost and have added additional features
such as in-circuit programmability and higher gate count. They tend to lend themselves
toward high-speed applications; real time video processing and Digital Signal Processing
(DSP). The CPLDs used for this design cost $50.00 a unit making them attractive for a
tight budget. Both CPLDs and FPGAs are available in SRAM based programming
configuration but only CPLDs can be EPROM or EEPROM programmed. This means
that CPLDs can be up and running when power is applied and are nonvolatile.

Some

FPGAs are antifuse-based allowing them to only be one time programmable devices
(OTPs) and SRAM based, which can be programmed as many times as required.
High-density PLDs come in two basic architectures.

Figure 4.2 shows the

connectivity differences between the two architectures. They are segmented-block based
and channel-array-based. Segmented-block architecture consists of a series of logic
arrays and I/O macrocells that are connected together with an interconnect matrix. CPLD
Examples of these are Advanced Micro Devices/Vantis MACH family and Altera
Corporations MAX family of devices. Segmented-block FPGAs have different means of
implementing logic functions. One way is through the use of SRAM based Look-up
tables (LUTs) and the other is through multiplexer-based logic elements that are typically
in antifused-based devices like Actel Corporations ACT1, ACT2 and ACT3. Antifusebased FPGAs have less resistance than SRAM based, which decreases delay times, but
are non-reprogrammable.

Nonsegmented-block or channel-array-based contain input

pins connected to logic array elements that have interconnect paths to macrocells which
in-turn are connected to the I/O pins. Some examples of channel-array-based FPGAs are

22

Xilinx Corporation Logic Cell Array (LCA), Atmels 6000 series and Lucent
technologies ORCA series. Channel-array-based devices are register rich, have many
I/Os and have programmable interconnects between the logic elements and the I/O
blocks. The SRAM based devices are more resistive than the antifuse-based and are
typically slower.
4 Input Pins

36

Interconnect
Matrix

Logic
Array

Macro
Cell
I/O
Pins

Macro
Cell

Input Pins

32

36

Logic
Array

Macro
Cell
I/O
Pins

Macro
Cell

32

Segmented

Logic
Array

Macro
Cell

I/O
Pins

Macro
Cell

I/O
Pins

Macro
Cell

I/O
Pins

Nonsegmented

Figure 4.2 Architectural Comparison Between Segmented and Nonsegmented

There has always been a debate over when to use a FPGA and when to use a
CPLD. CPLDs are ideal for control circuits and state-machine based control logic. They
have fast, predictable timing. It is very difficult to predict the data path delays in a FPGA.
The greatest advantage to FPGAs is having finer logic blocks and a very flexible
architecture for implementation in control logic designs, data path designs, arithmetic and
logic functions as well as register rich designs.

23

4.3

Simple and Complex Programmable Logic Devices

4.3.1

Detailed Architecture
Low density PLDs consist of PAL or PLA structures. These structures can be in

the form of AND-OR, NAND-NAND or NOR-NOR in multi levels of logic with I/O
macrocells and I/O pins. Figure 4.3 shows a portion of a simple PAL architecture. The
four most important elements of the PAL structure are as follows:
1. The AND-plane provides the connections between the inputs and the AND gates or
the product terms that implement the logic functions or control logic.
2. The OR-plane makes the connections between the AND-plane and the outputs. The
OR-plane defines the number of ORed outputs, the number of product terms per
output, how the product terms are distributed over the outputs and the connections of
the Ored outputs with the storage elements.
3. The storage elements determine the structure of the outputs. The storage elements are
flip-flops that can provide clocking or feedback into the logic. The outputs can be
configured to be sequential or combinatorial and have active high or low pull up
resistors. The storage elements can be configured as edge-triggered D-type, J-K, R-S
or T flip-flops, or transparent latches.
4. The I/O pins can be configured as an input, output or bidirectional.
SPLDs of today are ten times faster then the original PALs. The propagation delay of
PLDs is down to 3.5 ns and even faster for state-of-the-art versions. The delay is slow
due to the oxide-isolated submicron 0.35 um processes and the UVEPROM and
EEPROM storage cells.

24

OR Array
(Fixed)

AND Array
(Programmable)
Q3

Q2

Q1

Q0

Figure 4.3 Simple PAL Architecture

CPLDs consist of logic blocks, programmable interconnects, and I/O cells. Each
logic block in the CPLD contains inputs, logic arrays and the array allocators, macrocells,
and I/O cells. Figure 4.4 contains a generic block diagram of the CPLD architecture and
functional description. The logic blocks perform sum of product logic function and store
the results in registers in the macrocells. The interconnections route signals to and from
the logic and I/O cells to implement the desired logic design.
The resources that are available in a CPLD are as follows:
1. The number of macrocells is the amount of registers available to implement the logic
function or the design to be implemented. The number of macrocells determines the
complexity of the design.

25

The Macrocell can be

Configured as Combinatorial or
Registered
Logic Block

Clock / Input
Pins

Interconnect
Matrix

Array
and
Allocator

Macro
Cell

I/O
Cells

I/O
Pins

Macro
Cell

Input Pins

The HighSpeed Matrix

Routes Signals
Throughout the
Device

The Logic Array

Consists of
Product Terms
That Form the
Basis of the
Logic Being
Implemented

The Allocator
I/O Cells
Distributes
Provide a
Product Terms
Three-state
for Efficient
Buffer
Logic
Configurable as
Implementation Input, Output
and
Bidirectional

Figure 4.4 CPLD Architectural Block Diagram

2. The number of inputs to individual logic blocks is another important resource. If the
design requires more inputs to a specific set of logic blocks than are available, then
multiple logic blocks can be used to account for the lack of available inputs.
3. The final resource is the total number and allocation of product terms. The total
number of product terms that the device can OR together defines the upper limit for
the amount of logic that can fit into the CPLD. Product term allocation is sometimes
more important than the total number of product terms. Some times the allocation
function needs to feedback through multiple logic cells to implement the logic
function at the expense of increased delay.

Sometimes the allocation function

requires that product terms be assigned to individual macrocells or product terms to

be assigned to multiple macrocells for product term steering and sharing respectively.

26

The CPLD performance is dependent upon the interconnections between logic cells
for which there are two different aspects as follows:
1. The number of signals that the global interconnect provides to each logic cell
determine how well the CPLD will perform on a specific logic design. If the amount
of input signal to the logic cells is smaller than the number that is required for the
logic function then it will have to be slit between multiple logic cells to account for
the lack of input signals.
2. The second performance aspect is routability efficiency or how well the interconnect
signals get routed between the logic cells in the most efficient manner. When the
number of signals that need to be routed reaches a maximum for the specific design
and chip architecture, it becomes more difficult to route the signals.
CPLDs consist of two types of interconnect structures.
1. The first type is called the crosspoint switch architecture.

This architecture is

completely routable because any combination of signals can be routed from any logic
block via the global interconnect. The only problem with this architecture is that the
vast amount of routing capability decreases the overall speed and increases the die
size.
2. The multiplexer approach is used instead of the crosspoint switch to increase the
overall speed and to reduce the size of the chip. Figure 4.5 contains a block diagram
of the multiplexer-based interconnect structure.

Many vendors have different

approaches to try to optimize the routability efficiency.

CPLDs have a wide input structure and large sum of product arrays. They are ideal
for implementing fast state machine designs, high speed control logic and fast decode

27

applications. Many vendors have implemented the in-system programmability function

that reduces on the chip handling and makes it flexible to change designs while the chip
is mounted on the board.

I/O

Logic
Block

Logic
Block

Logic
Block

Logic
Block

Logic
Block

Multiplexer
Based
Switching
Network

Logic
Block

I/O

Logic
Block
Logic
Block

Figure 4.5 CPLD Multiplexer-Based Interconnect Structure

4.3.2

Programming Technologies
The programming technology of earlier PLDs was based on bipolar processes,

which came from PROM technologies. The PROM structure contained lateral fuses and
vertical fuses. The lateral fuses, or metal alloy, would interconnect two wires. When the
lateral fuses are blown due to the current through a bipolar transistor, the two wires are
not connected. The vertical fuses did not have the metal alloy but they did have the
bipolar transistors in series with back-to-back diodes. The diodes would isolate the two
wires in the not blown state. Enough current through the transistor would cause an
avalanche effect that would short one of the diodes connecting the two wires. The
fuses are made out of different materials like nichrome, titanium-tungsten and platinum
silicide.

The fusible-link PLDs have different programming requirements that are

specified by the vendor in the data sheets.

28

The programming technologies for CPLDs are UVEPROM and EEPROM. In the
EEPROM and UVPROM CPLDs transistors are used as switches to make interconnects
just as the fuses in the SPLDs. The programming technology is an important element
when selecting devices. The selection of a particular device also depends on volatility,
nonvolatility, reprogrammability, in-system programmability (ISP), program time, erase
time and testability.
The electrically erasable CMOS technology is based on a stored charge on
a transistor with a floating gate or a gate that has no connection. There are actually two
transistors, one that has the floating gate and another that is the control gate. In the
programmed state the transistor is turned on due to a positive charge caused by a deficit
of electrons in the floating gate. Figure 4.6 shows the schematic for the electrically
erasable programming cell.

Read Line

Program Line

Word Line

Program Transistor

Read Transistor

Electron Source
Tunnel Oxide
Control Gate
Sense Transistor
Floating
Gate
Source

Figure 4.6 Schematic of Electrically Erasable Programming Cell

29

Hot electron injection and tunneling are the two ways of transferring a charge on
to the floating gate.

UVEPROM uses hot electron injection and EEPROM uses

tunneling. In hot electron injection a bias is set between the source and the drain of the
transistor of the control gate and the substrate. A strong current is present causing the
channel to be pinched off. A field is formed in the diagonal direction by two applied
electrical fields. The electrons can not pass in that direction due to the oxide barrier.
Some electrons have enough energy to cross the barrier at the shortest point over into the
floating gate where they are trapped. The term hot electron injection refers to the high
electron fields. During programming large electron fields are established to get a large
amount of electrons over the oxide barrier and over to the floating gate. Exposing the
device to UV light will erase the device because it causes the electrons to cross back over
the barrier. The UVEPROM devices need a window to allow the UV light to enter the
device. The special packaging of the device increases the cost and results in about 30 to
45 minutes to erase the device. Figure 2.7 shows a schematic of hot-electron injection in
UV-erasable devices.
Electrically erasable devices use Fowler-Nordheim tunneling to get the electron
over the barrier onto the floating gate. The electrons tunnel through the barrier when a
field is applied. For electrically erasable parts the oxide barrier is about one-third the
thickness as it is for UV parts, making the field much less to get the electrons across.
EEPROM based devices are more cost effective than UVEPROM based devices.
To improve the performance of electrically erasable PLDs, the programming cell
has been divided into two parts. One is the programming portion and the other is the data
path portion.

The programming portion requires transistors that can handle high

30

electrical fields and the data path portion requires transistors that are fast. A cell is
programmed if it generated a negative voltage as a result of no charge on the floating
gate. Large amounts of charge on a floating gate caused by applying a negative voltage
will erase a cell. When a voltage is applied between the program and control gate nodes,
the direction of that voltage determines if the device is getting programmed or erased. If
the control gate is given a positive voltage and the program node is grounded the device
is being erased. The positive voltage draws electrons across the tunnel oxide form the
program transistor to the floating gate, which turns the read transistor OFF.

For

programming, the program node voltage is higher and the control gate is grounded
causing electrons to flow from the gate leaving behind a positive charge that turns the
transistor on.

Control Gate

Floating Gate
Hot Electron Injection

Oxide Barrier
Field
e-

eee-

e-

n+source

n+drain
Field
Pinch-Off Channel

Depletion Region

P-substrate

Figure 4.7 Schematic of Hot-Electron Injection

31

4.3.3

Simple and Complex Programmable Logic Devices Summary

There are many different SPLD/CPLD vendor and their device families and

architectures. The major SPLD/CPLD vendors and device families are AMD/Vantis with
PAL and MACH families, Altera with MAX and FLEX families, Cypress Semiconductor
EPLD/CPLD families, ICT Peel arrays family, Phillips Semiconductors CPLD families
and Xilinx FastFlash CPLD family, just to name a few. Depending on the design
requirements and specifications, there are tradeoffs on selecting FPGA architectures over
CPLD architectures. There are also tradeoffs in selecting a specific CPLD architecture to
meet the system requirements.

The next section will cover the CPLD vendor and

architecture chosen to implement the design. Lattice Semiconductor is the CPLD vendor
with high-density in-system programmable Large Scale Integrated (ispLSI) CPLDs. The
Lattice CPLD architecture provides very flexible architecture with high-density logic.
The Camera design implemented is for a research and development project with general
requirements.

The in-system programmable feature allowed for ease in design

modifications and debugging capability. The cost of the devices and the Electronic
Design Automation tools to implement the design was reasonable for the budget
available. A very important feature is the predictable timing of a CPLD. The interface
design was a complex design done in VHDL.

The predictable timing allowed for

learning the design and the code rather than the device architecture and timing issues.
The Lattice devices also have a wide range of speed options, which meet the
requirements of the design. The VHDL implementation of the design will be discussed
in Chapter 6, VHDL Design Entities.

32

4.3.4

Lattice Semiconductor Overview

Lattice Semiconductor CPLDs are nonvolatile EECMOS technology. There are

six families and they are 1000/1000E, 2000/2000V, 3000 and 6000. The densities range
from 1000 to 25,000 PLD gates. A very important feature of this architecture is the
Global Routing Pool, which connects all the internal structures and I/Os with predictable
timing. Another key element is the Generic Logic Blocks (GLB). They provide a high
input to output ratio for design optimization. The architecture also allows for a ProductTerm-Sharing Array (PTSA) and two types of clocking signals. The two clocks are an
external globally distributed snychronous clock and an internal globally distributed
asynchronous product term clock. The Output Routing Pool (ORP) allows for flexible
interconnections between the GLBs and the I/O modules.
4.3.4.1

Lattice Semiconductor ispLSI 1000/E

The GLBs for the 1000/E contain 18 inputs, programmable AND/OR/XOR array

and four outputs. The four outputs can be configured as combinatorial and registered.
The difference in the 1000E is that it has two more global output enable pins and a
programmable output slew rate control.
The Global Routing Pool (GRP) is the internal routing structure that connects the
GLBs to the I/Os and the I/Os to the GLBs. The output of any GLB is available for input
into any other GLB. The I/O pins are also available to interconnect to any GLB input.
The ORP is the routing structure in the architecture that allows the GLBs to be connected
to the I/Os. This allows flexibility when assigning I/O pins.

A megablock consists of

eight GLBs and 16 I/O cells. The 1000E has two dedicated input signals that are common
to all GLBs and the product term OE which is common to all 16 I/Os. The densities in

33

the 1000/E family range form 16 GLBs to 48 GLBs. Figure 4.8 contains a functional
block diagram of one of the ispLSI 1000 devices.

I/OI/OI/OI/O I/OI/OI/OI/O I/OI/OI/OI/O I/OI/OI/OI/O

Input Bus

Generic
Logic Blocks
(GLB)

IN

Output Routing Pool

Input Bus

Output Routing Pool

GLB GLB GLB GLB GLB GLB GLB GLB

GLB GLB GLB GLB GLB GLB GLB GLB

IN IN

I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O

Clock
Network

Input Bus

Output Routing Pool

GLB GLB GLB GLB GLB GLB GLB GLB

Global
Routing
Pool
(GRP)

IN

GLB GLB GLB GLB GLB GLB GLB GLB

Megablock

I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O

IN IN

CLK0
CLK1
IOCLK0
IOCLK1

Output Routing Pool

Input Bus

Y3 Y2 Y1 Y0

I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O I/O

IN IN

Figure4.8: isp LSI Functional Block Diagram

A GLB is divided into four sections. The four sections are the AND array,
Product-Term-Sharing Array (PTSA), reconfigurable registers and the control functions.
The AND array in the PTSA consists of 20 product terms that has 18 inputs available to
produce a logic sum. 16 of the 18 inputs are from the GLP that can be connected to I/Os
or any other GLB. The other two inputs are the dedicated inputs. The PTSA routes the
20 product terms to the 4 GLB outputs. The OR gates have four, five, or seven product
term inputs to them. The OR gates can be routed to any GLB output. The PTSA allows
34

for product term sharing or combining. The reconfigurable registers are XOR gate inputs
to D-type flip-flops. The XOR gate can be used as a logic element or to configure the
register as a JK or T-type flip-flop. The registers can be bipassed if the design requires
combinatorial logic and is brought out to the GRP. The I/O signals are routed to the
GRP. The I/O cell is used to route input, output and bidirectional signals connected to
the I/O. There are five global clocks for the devices. CLK0, CLK1 and CLK2 are for
GLB clocking and IOCLK0 and IOCLK1 are for clocking the I/O cells.
4.3.4.2

Lattice Semiconductor ispLSI 2000/V

The ispLSI 2000 family has densities ranging from 32 macrocells up to 128

macrocells, which is equivalent to 1000 PLD gates and 6000 PLD gates respectively.
The 2000V device is a low voltage version of the 2000 at 3.3 V. The global logic block
(GLB) for the 2000 family is essentially the same as for the 1000 family. Figure 2.9 is
the schematic of the generic logic block for the 2000 family. The general architecture has
the following modifications and enhancements:

GLB and clock multiplexer scheme is different on the 2000 series. There are three
dedicated clocks CLK0, CLK1 and CLK2. These clocks are used to clock all the
GLBs that are configured as registers. The clocks are used inside the GLBs and are
selected by a multiplexor. The internal clock signals are connected to the I/O pins Y0
through Y2.

The 2000/V family made a simplified I/O register. The I/O cell can be configured as
an input, output and bidirectional with tri-state control. The I/O clocks were removed
and the outputs can be configured as totem pole or open-drain.

35

The Output Routing Pool (ORP) was enhanced for further routing capability. The
2000 series devices have two sets of ORPs per megablock allowing all 32 outputs of
the GLBs to have access to the 32 I/O cells. This architecture doubles the routability
of this family.

Inputs From Global

Routing Pool
0

1 2

4 5

Dedicated
Inputs

Product Term
Sharing Array 3 PTs
(PTSA)
and

8 9 10 11 12 13 14 15 16 17

D Registers

XOR

0
1
2
3
4
5
6

4 PT
Bypass

7
8
9
10
11
12
13
14
15
16
17

M
U
X

M
U
X

M
U
X

M
U
X

Q3

Q2

Single PT

7+4 PTs

Q1

Q0

18
19

AND Array

PT Reset
Global Reset

Control
Functions

PT Clock
CLK 0
CLK 1
CLK 2

MUX

To Output Enable Mux

PT Output
Enable

Figure 4.9 ispLSI 2000/V Generic Logic Block

4.3.4.3

Lattice Semiconductor ispLSI 3000/V

The ispLSI 3000 family of devices range from 7000 PLD gates / 192 macrocells

to 14,000 PLD gates / 320 macrocells. The basic architecture of the device is similar to
the 1000 series with the following exceptions:

36

The generic logic blocks of the 3000 series devices are called twin GLBs. A twin
GLB contains 24 inputs and 8 outputs. The GLB is divided into four sections and
they are the AND array, PTSA, reconfigurable registers and the control section. The
AND array consists of 20 PTSAs that can accept all 24 twin GLB inputs. The GLB
inputs come from the I/Os or other GLBs. There are two sets of PTSAs in the twin
generic logic block. There are 4 OR gates in the PTSA that have 4, 5 and 7 inputs
from the AND array. The output of the four OR gates can be routed to any of the four
outputs on the twin GLBs if more product terms are needed.

A megablock in the 3000 devices consists of four twin GLBs with 24 inputs per twin
GLB. There are no dedicated inputs. There are single-I/O devices and double-I/O
devices. In the single-I/O structure, the 32 I/Os from the 4 twin GLBs get routed to
only 16 I/Os through the ORP. There is only one I/O pin for every two GLBs. If
your design contains many I/Os, this could cause a problem.

The double-I/O

structure allows for 32 I/O pins for all 32 GLB output pins. Figure 2.10 is the Lattice
ispLSI 3000 family megablock layout for double-I/O devices.

The global clock structure was changed form the 1000 family. There are now a total
of 5 clocks that are networked to each GLB. Three clocks are for register clocking
and the remaining two are for I/O clocking.

The I/O cells are the same as the 1000/E families with the exception of the added
boundary scan capability. There are added boundary scan registers that allow for
JTAG IEEE 1149.1 on-chip test capability during the PCB testing. There are also
added global OE pins for static testing.

37

Global Routing Pool (GRP)

24

24

Twin GLB A0

24

Twin GLB A1

Twin GLB A2

Output Routing Pool

24

16

Twin GLB A3

Output Routing Pool

I/O Cells Inputs to GRP

IO IO IO IO IO IO IO IO

IO IO IO IO IO IO IO IO IO IO IO IO IO IO IO IO

IO IO IO IO IO IO IO IO

32 I/O Cells
Figure 4.10 ispLSI 3000 Megablock Layout
4.3.4.4

Lattice Semiconductor ispLSI 6000

The ispLSI 6000 family of devices contains high-density cell-based

programmable logic, dedicated memory module, a dedicated register / counter module

and an 8000-gate area for general purpose logic. There are three devices in the family,
which contain different types of memory modules. The ispLSI 6192FF contain FIFO, the
ispLSI 6192SM contain single-port RAM and the ispLSI 6192DM contain duel port
RAM. The basic architectural elements of the 6000 family are as follows:

The twin Generic Logic Block (GLB) is the basic structure for the 6000 family. In
the ispLSI 6192 there are 24 twin GLBs each containing 24 inputs, programmable
AND, 2 OR / exclusive-OR arrays and 8 output blocks.

The outputs can be

configured as combinatorial or registered.

The memory module can be configured as a dual-port FIFO, single-port RAM or

dual-port RAM. The FIFO has dedicated input signals, its own control flags and can

38

be programmed as a 256 x 18 or a 512 x 9. The single-port RAM can be configured

as a single 256 x 18 or 512 x 9 or two smaller 128 x 8 or 256 x 8 memories. The duel
port can be configured as a 256 x 18 or 512 x 9 and can read and write concurrently.

Eight 16-bit blocks can be configured as registers or shift registers. Four of them can
be programmed as loadable up/down counters.

The ispLSI 6000 devices also include boundary scan capability.

4.3.4.5

Lattice Semiconductor Summary

Lattice Semiconductor has recently introduced two new families into the high-

density CPLD profile and they are the 5000 and 8000 family. Both these families have
wider input capability into the GLBs and higher density of gates. The 8000 family has up
to 45,000 PLD gates. The ispLSI 3320 70LM 208 was the device chosen for the design.
It contains 320 macrocells, 14,000 PLD gates, 162 I/Os and supports a maximum
frequency of 100 MHz. It allowed for high-density designs, flexible partitioning of the
design and flexible development and debugging using the in-system programmable
feature. This device was available in a 208-pin Metal Quad Flat Pack (MQFP), which
made it ideal for prototyping the electronics without getting into Ball Grid Array (BGA)
packaging issues. BGA Packaged devices have the I/O pins very close together and do
not interface with standard Pin Grid Array (PGA) carriers.

BGA are very small

packages, ideal for Printed Circuit Board (PCB) applications.

4.4

Field Programmable Gate Arrays

4.4.1

Introduction
When considering an FPGA architecture, one must look at two main concerns;

size / flexibility of the logic cell and the routing architecture. The logic cell can be as

39

simple as a transistor or as complex as a microprocessor. The logic blocks can be built

from transistor pairs, small gates, multipelxers, look-up tables or wide fan-in AND-OR
structures.
FPGA logic blocks differ in size and capabilities. Some FPGA vendors employ
the sea-of-tiles architecture, which are small configurable units as logic cells and other
vendors like Xilinx use the lookup table architecture which is relatively large. The logic
cells can be defined by their granularity. Granularity is defined as the number of boolean
functions, number of two-input NAND gates, total number of transistors or the number of
I/Os. Granularity is broken down as follows:
1. Fine-Grain Logic Cells Fine grain logic cells consist of a few transistors. The
advantage of fine grain logic cells is it is easier and more efficient to implement
complex designs in transistor level logic cells. The disadvantage, is that fine grain
logic cells require a more complex and larger programmable interconnect structure
increasing the size of the chip as well as internal chip delays.
2. Course-Grain Logic Cells An example of a course-grain logic cell is the multiplexer
architecture. The advantage of course-grain logic cells is that they can implement
complex functions with very few transistors internal to the logic cell.

The

disadvantage it that a large number of inputs will be required for complex logic
functions which stresses the influence of routing resources. Course-grain logic cells
are associated with antifuse programming technology.
The routing architecture of an FPGA is the internal interconnection for the logic cells
implementing a logic function. The interconnection resources consist of the connection
block and the switch block. The connection block provides the interconnection between

40

the horizontal and vertical paths and the logic cells and the switch block provide the
connections in the horizontal and vertical paths. The routing architecture affects density
and performance of the FPGA.
routability and speed.

The routing architecture must meet two criteria:

Routability is the ability of the FPGA to provide all the

interconnections for a particular design. The speed of the FPGA is closely related to the
propagation delay through the routing. This delay is 40% - 60% of the total delay. The
programmable switch also introduces a RC delay to the total delay.
FPGAs are divided into three different architectural groups: (1) row-based (2)
symmetrical and (3) fine-grain cellular. Row-based FPGAs are usually course-grain
logic cells, organized in a row and divided horizontally by interconnect signals. The
symmetrical architecture for FPGAs consists of rows and columns of large grain blocks
called configurable logic blocks (CLBs) with horizontal and vertical routing channels.
The fine-grain or cellular architecture can be described as the sea-of-gates approach in
which the logic cells are very small or fine. The next section will describe these in
detail.
4.4.2 FPGA Architectural Groups
4.4.2.1

Row-based Architecture
Row-based FPGAs contain rows of logic cells that are separated by

programmable interconnect structure or routing channels. Actel is a perfect example of a

row-based FPGA. Figure 4.11 shows a very simple schematic for row-based FPGA
architecture. Row-based FPGAs contain different lengths of wire segments that are
connected together using antifuses.

41

Horizontal Interconnect /
Routing Channel

Logic Cells

Figure 4.11 Schematic of Row-based FPGA

A critical design issue with respect to row-based FPGAs is having an accurate

estimate of the number of tracks required to implement the design with respect to the
number of tracks available in the chip. Typical designs could have tracks that run the
length and the width of the chip consisting of many wire segments of varying length.
Minimizing the number of fuses for any given net to increase the performance and timing
characteristics is another critical issue with respect to row based FPGAs. There have
been routability analysis and routing algorithms developed for the optimization of the
routing channels in row-based FPGAs. The segmentation of the routing channels can be
handled as unconstrained routing channel, fully segmented block channeling, nonsegmented channeling, 1-segment channeling and 2-segment channeling. A detailed

42

description of each channel architecture will not be discussed but the results of the
analysis are as follows:

The wire segment length that connects the logic cells affects the performance of the
design.

It is critical to reduce the amount of switches or fuses to reduce the RC delay factor.

It is important to understand the use of the routing design tools with respect to vendor
devices. The tools take into account speed and performance as well as the area of
routing channels available for specific chip architectures. It is important for user
interaction to define critical nets to reduce the amount of tracks to meet performance
requirements.

4.4.2.2

Symmetrical (SRAM-based) FPGA Architecture

Symmetrical or SRAM-based FPGAs contain memory cells as the logic blocks to

implement the logic functions to create a design. There is no RAM area in the FPGA so
it is distributed among the logic cells. SRAM FPGAs are built for density and stability
and not necessarily for speed. The main disadvantage of SRAM-based is that they are
volatile and need to be reprogrammed once power is reapplied. Xilinx is an excellent
example of an SRAM-based FPGA with its architecture based on Logic Cell Arrays
(LCAs) which contain functions as Look-Up Tables (LUTs). Figure 4.12 contains a
simple schematic of an SRAM-based FPGA.
The RC delay introduced by each Programmable Interconnect Point (PIP) adds up
over the routing path. The fewer PIPs used in SRAM based FPGAs will decrease the
loading in the interconnect segments but reduce the routability. The more PIPs improve

43

routability but increase loading because with each PIP there is a memory cell that can
increase the die size of the FPGA.
Interconnect Structure
Logic Blocks

I/O Cells

Figure 4.12 Schematic of Symmetrical / SRAM-based FPGA

SRAM-based FPGAs consist of three building blocks. The first is the memory
cells that implement the logic function. The second is the Programmable Interconnect
Point (PIP) and the third is the multiplexer controlled by the memory cell. The PIP is a
pass transistor that is on or off depending upon the value in the memory cell. The pass
transistor introduces a RC delay. The multiplexer is a uni-directional routing structure
that can be as wide as necessary to implement the function.
In an FPGA architecture there are trade-offs between the size of the logic block
and the speed and performance at which the design is implemented. Small logic blocks

44

have their advantage but may lead to slower performance due to the many interconnects
that are required to implement a function. Large logic blocks may be able to handle
complex designs while decreasing the amount of interconnects but have poor logic
efficiency. The capacity and performance are dependent upon the architecture of the
FPGA and the software tools that provide placement, routing and partitioning of the logic
functions.
Recently FPGA vendors such as Actel, Altera and Xilinx have incorperated
medium-size logic blocks of memory such as single-port or multi-port SRAMs, FIFOs
and buffers. The imbedded memory allows designers to simplify designs and have the
design operate faster with access time to the memory modules ranging from 5 to 20 ns.
Xilinx used the LUT configuration in small blocks of RAM that is distributed throughout
the chip where Altera and Actel used dedicated blocks of RAM within the FPGA.
4.4.2.3

Fine-grain (or cellular) FPGAs

Fine-grain FPGA structure consists of small logic cells in a compact structure for

direct interconnect to perform local logic functions. Motorola Programmable Array

(MPA) is a perfect example of the fine-grain architecture. The interconnect structure is
hierarchical consisting of fast connections, medium interconnect and global interconnect.
The local connections are very fast and join cells at the lowest level to form the
macrocells that can be used to implement small functions like counters and comparators.
The medium interconnect combines the macrocells together and the global interconnect is
for the complete routing and chip layout.
The functionality of a cell is different for fine-grain FPGAs than course-grained
channel-array devices. A basic cell is an AND / XOR function or an AND / D-type flip-

45

flop set. If this logic cell was to be used in a channel arrayed device, the cell would be
very inefficient due to the routing resources. The hierarchical routing and the basic logic
cells combined into macrocells make fine-grain devices very efficient with very low
routing overhead. There are two advantages to using fine-grained FPGAs and they are as
follows:
1. The logic cells are small enough to implement simple logic functions without having
wasted resources.

This reduces redundancy when implementing simple logic

functions.
2. Since there are few critical paths through the cell, the cell can be optimized.
4.4.3
4.4.3.1

FPGA Programming Technologies

Antifuse
An antifuse is a two terminal device that when unprogrammed has a very high

resistance between the two terminals and when programmed, or blown, creates a very
low resistance or permanent connection. The application of a high voltage from 11 V to
21 V will create the low resistive permanent connection. Antifuse technologies come in
two types. The first is oxide-nitride-oxide (ONO) dielectric based and the other is
amorphous silicon or metal-to-metal antifuse structures.
Dielectric based antifuses consist of a dielectric material between N+ diffusion
and polysilicon which breaks down when a high voltage is applied. Early dielectrics
were a single-layered oxide dielectric until Actel came out with the programmable low
impedance circuit element (PLICE), which is a multi-layer oxide-nitride-oxide (ONO)
dielectric fuse.

A high voltage across the PLICE melts the dielectric and creates

polycrystalline silicon between the terminals. When the PLICE is blown, it adds three

46

layers rather than the double metal CMOS process. The layers are a thin layer of oxide
an top off the N+ surface, Low-pressure Chemical Vapor Deposition (LPCVD) nitride
and the reoxidized top oxide. The programming current has an important effect because
the higher the current during programming, the lower the link resistance, resulting in
smaller thickness for the antifuse material. Programming circuits for antifuses need to
supply high currents (15 ma for Actel) to insure high reliability and performance.
Amorphous silicon antifuse technology is the alternative to dielectric antifuse. It
consists of amorphous silicon between two layers of metal that changes phases when
current is applied. When the antifuse is not programmed the amorphous silicon has a
resistance of 1 Gohm. When a high current (about 20 mA) is applied to the anitfuse the
amorphous silicon changes into a conductive polysilicon link. QuickLogic pASIC FPGA
is a perfect example of an amorphous silicon antifuse technology.
4.4.3.2

SRAM-based
SRAM FPGA architecture was discussed in section 4.4.2.2 and consists of static

RAM cells to control pass gates or multiplexers. The FPGA speed is determined by the
delay introduced by the logic cells and the routing channels. Multiplexers, look-up tables
and output drivers affect the speed of signals through the logic cells. An FPGA with
more PIPs is easier to route but introducing more routing delay. The size of the look-up
table plays an important role depending on the design. Smaller LUTs provide higher
density but larger ones are preferred for high-speed applications.
4.4.3.3

SRAM vs. Antifuse

The following is a list of advantages and disadvantages for the two technologies:

47

1. Antifuse programming technology is faster than SRAM programming technology due

to the RC delays introduced by the interconnect structure.
2. Antifuse technology has more silicon area per gate and is easier to route than SRAM
technology.
3. A disadvantage of antifuse FPGA is that they require more process layers and mask
steps and also contain high voltage programming transistors.
4. SRAM-based technology contains higher capacity than antifuse technologies.
5. SRAM based technology is very flexible with in-system programmability and the
ability to reconfigure the design during the debugging stage while antifuse technology
is one-time programmable (OTP). This ability reduces design and development,
which reduces overall cost of the design. Another advantage to this is that SRAM
technology can be programmed at the factory through complete verification test
where the antifuse are tested as blanks and require programming by the user to
verify design requirements and operation.
6. A disadvantage of SRAM technology is that it is volatile meaning it has to be
reprogrammed every time power is turned off and on again. The SRAM usually
require an extra memory element to program the chip which occupies board space
4.5

Programmable Logic devices for Space Applications

4.5.1

Introduction
Over the years programmable logic devices have been used more and more in

space and military applications because of there design flexibility, chip density for large
complex designs, chip cost and fast turn-around times. ASICs and gate arrays have been
used in the past and are still being used due to high performance and greater radiation

48

tolerances but they have high nonrecurring engineering (NRE) costs and they have longer
turn-around times for engineering model and flight units.
4.5.2

Space Environment
The space environment has radiation risks that affect electronic components and

devices as well as programmable logic devices on earth orbiting spacecraft missions and
deep space missions. The space environment consists of radiating particles like highenergy electrons, protons, alpha particles and cosmic rays. These irradiating particles in
the space environment, can cause problems and latch-ups in the electronic circuits.
Electronic components in space are not affected by high doses of radiation but over a
period of many years, depending on the mission, electronic system can accumulate a
significant amount.
Ionization is high energy creating electron-hole (h-e) pairs when passing through
a material. Ionization is the dominant effect of charged particles in space. Radiation can
break atomic bonds by displacing the crystal lattice structure in a material and cause
trapping recombination centers. Both of these can lead to problems in the performance of
the device.
The electromagnetic radiation types that cause ionization are X rays and gamma
rays. Gamma rays affect the electronic component by the charge interacting with the
material. Ionizing particulate radiation can be in the form of uncharged particles, light
charged particles and heavy charged particles. Neutrons are uncharged particles.
Uncharged particles like neutrons cause crystal lattice damage. Electrons, protons, alpha
particles and beta particles are charged particles. Lightly charged particles interact with
the material of the device just as gamma rays. Iron, bromine, krypton, and Xenon are

49

examples of heavy charged particles. Heavy charged particles are in cosmic rays. The
sun and galactic sources cause cosmic rays. Heavy charged particles create volume
ionization within the substrate of the electronic component. The charge form heavy
charged particles in cosmic rays collect on circuit nodes and causes problems in data loss
through single-event upsets (SEUs) and transient upsets. Table 4.2 lists radiation types
and the related failures. Radiation damage caused by charged particles form space
environments can be broken down into total ionizing dose (TID) and single-event
phenomenon (SEP) or single-event effect.
4.5.3
4.5.3.1

Radiation Damage
Total Ionizing Dose (TID)
Bipolar technology devices are not as sensitive with respect to total ionizing dose

effects as MOS technology devices. Major radiation damage is caused by charged

particles built up in passive oxides surrounding silicon surfaces. Bipolar technology
depends on transistor junction for operation where MOS technology depends on surface
and interface signals.
PLD designers prefer MOS technology because of low noise and power. It is
sensitive to total ionizing dose effects. Trapped charge in the oxide layer of the CMOS
device is the radiation damage related to ionization effects. The silicon dioxide (SiO2) is
the most sensitive to ionizing radiation. As the radiation passes through the silicon
dioxide, it creates electron-hole pairs. The field in the silicon oxide allows the electronsholes to drift apart. The electrons drift out of the oxide and are collected by the gate
electrode. Some of the holes form oxide traps and others form interface traps. Oxide
traps stay where they were generated having a positive charge and the interface traps are

50

energy levels in the SiO2/Si interface. Interface traps reduces the conductance of the
channel of the MOS transistor. The oxide traps and interface traps accumulate over time.
The total ionizing dose received by a programmable logic device is expressed in rads (Si)
or rads (siO2) as follows:
1 rad (Si) = 100 ergs absorbed / gram of Si
As a reference, Table 4.3 shows various levels of radiation expressed in rad (Si) for
specific application requiring qualified parts.
Type of
Radiation
Total Dose
(Low dose rate)
(High total dose)
Transient
(High dose rate)
(Low total dose)
Electrons and
Protons

Type of Degradation Failure

Transistor Level
Photons (X rays and gamma rays
Threshold voltage shifts
Drive current shifts
Carrier mobility reduction
Parasitic leakage current
Collapse of depletion region
Burn-out of metal interconnects
Damage to junction region
Charged particles and cosmic rays
Threshold voltage shifts
Drive current shifts
Carrier mobility reduction
Parasitic leakage current

Alpha particles

Carrier recombination and trapping

Collapse of depletion region (photons
only)
Collapse of depletion region

Heavy ions

Collapse of depletion region

Neutrons

Uncharged particles
Junction leakage
Bipolar beta reduction
Carrier recombination and trapping

Circuit Level
Performance degradation
I/O parametric shift
Lost functionality
Data upset/loss
Latchup
Catastrophic failure
I/O glitches
High photocurrents
Performance degradation or
improvement
I/O parametric shifts
Lost functionality
Data upset (single bit/node,
protons only)

Data upset (single bit/node,

protons only)
Data upset (single or
multibit/node)
Hard oxide failures
Latchup
No performance effects on
CMOS
Improves latchup tolerance on
CMOS circuits
Performance degradation on
bipolar circuits

Table 4.2 Types of Radiation and Related Failures

51

Total Dose, rads

100
1K
10K
100K
(Si)
Nuclear
xxxxxxxxx xxxxxxxxx
Explosion
Spacecraft
x xxxxxxxxx xxxxxxxxx
Satellite
xxx xxxxxxxxx xxxxxxxxx
Aircraft
xxxxxxxxx
xxxx
Ground-based
xxxxxxxxx
xx
Systems
Human Exposure Serious
Usually
Consequence
illness
fatal
Note: The xs in the table indicate a range within the total dose columns.

1M
xxxxxxxxx
xxxxxxxxx

Table 4.3 Ranges of Total Dose Exposures for Various Environments

4.5.3.2

Single-event Phenomenon
Single-event phenomenon (SEP) refers to the effect of high-energy particles

hitting the sensitive nodes in the electronic component. They can be in three forms and
they are single-event upset (SEU), single-event latch-up (SEL) and transient radiation
effects.
1. Single-event upset (SEU) is the affect of charged particles causing logic change in
flip-flops and RAM configuration bits in programmable logic devices. An SEU is
called a soft error because the charged particles collect on a sensitive bit node in a
memory element cause it to change logic level. An SEU is temporary and will regain
operation upon system reset.
2. Single-event latch-ups (SEL) are called hard errors because when they happen the
device can not recover from them.

They are either low impedance states or

avalanching in the drain junction of MOS transistors in CMOS devices. The hard
errors are caused by single charged particles that induce photocurrent that can be at
high levels causing burnout in the junction of the transistor.

52

3. Transient Radiation effect deals with the rate of the ionizing dose rather than the total
ionizing dose effect. A higher rate of ionizing radiation causes a greater number of eh pairs to be caught in the insulating layer and silicon. Transient upset is caused by a
burst of ionizing radiation rather than from one charged particle.
As there is an increase in the charged particles from ionizing radiation, there is a
loss of energy. This loss of energy is a result of the amount of energy mass and the
amount of charge from the ionizing particles. The Linear Energy Transfer (LET) is the
rate of energy from the charged particles (dE/dX) and is measured in units of
MeV/[cm^2/mg] or MeV/um. The lower the LET the more chance the device will have a
SEU or SEL. Proton belts have energies above 30 MeV, which is the minimum energy
that can penetrate a spacecraft. 8 MeV is the minimum level for proton induced upsets.
LET is found experimentally by testing the device under ionizing radiation. The LET is
directly related to the specific design or the technology used. The critical charge (Qc) is
the minimum charge necessary to cause a single-event phenomenon and can be simulated
in a design.
Earth orbiting satellites operate in low earth orbit or geosynchronous orbit, which
are 150 to 600 km and 35,880 km respectively. The magnetosphere, which is 6,380 km
contains the highest radiation around the earth. The magnetosphere contains heavy
charged particles and gets hit by heavy ions from solar and galactic cosmic rays. The
interplanetary spacecraft encounter heavy charged particles form other planets
magnetospheres and heavy ions from cosmic rays.
Electronic designs using PLD for space applications have to be concerned about
the following items:

53

Protons and electrons trapped in Van Allen radiation belts

The Van Allen radiation belts are two concentric donut-shaped clouds that store and
trap charged particles form solar wind. They are aligned with the magnetic axis of
the earth. Satellites and spacecraft passing through this region are bombarded with
protons and electrons. They are 200 km to 1200 km from the earths surface.

Heavy ions in the magnetosphere

Cosmic ray protons and heavy ions

Protons and heavy ions form solar flares

Altitudes of about 1 to 12 Earth radii (where Re = 6380 km) contain electrons with

energy levels of about 7 MeV. Protons from the solar and galactic rays have energy
levels ranging from 1 Mev to 800 MeV. The actual radiation experienced by a spacecraft
depends on the orbit, mission launch time and duration of the mission. A second source
of radiation is the charged particles found in radiation belts. Besides the energy from the
radiation belts there is sets of radiation from solar flares and cosmic rays. The charged
particles from solar flares can last from a few hours to several days. Solar flares contain
protons, electrons, alpha particles and heavy ions. Cosmic rays have energy ranging from
a few million electron volts to 10 GeV and are composed of mostly protons, some alpha
particles and heavy ions. There is a geomagnetic field around the equator that shields
spacecraft from charged particles when orbiting at low altitudes and low inclination. The
U.S. space station is in that type of orbit. The higher the altitude and larger the inclination
the more solar flares and cosmic rays. If the inclination of the orbit reaches the polar area
of the earth there is no protection form solar flares or cosmic rays. The highly elliptical
orbits (HEO) of spacecraft have an altitude of 30,000 km and pass through the Van Allen

54

belt and get hit with constant cosmic rays and solar flares. Spacecraft in geostationary
orbits (GEOs) are exposed constantly to cosmic rays and solar flares. The total radiation
effecting the electronic components is not only a factor of radiation but also a factor of
the shielding of the electronics.
4.5.4

Radiation hard / Radiation Tolerant

An electronic component has to go through many test to determine it if can

withstand specific radiation levels. A Radiation Hard (RadHard) device is designed to

withstand levels of radiation that include and exceed 300 Krad. A device that is radiation
tolerant has not been designed or tested for radiation hardness but to some level of
radiation resistance. The hardness levels of a device that is radiation tolerant may not be
guaranteed and the process is less expensive indicating there may be important
parameters such as oxide thickness not properly monitored, specified or tested.
4.5.5

Radiation Hardness Assurance (RHA) Testing

In unhardened programmable logic devices the performance characteristics and

fabrication process vary form lot to lot within vendor parts. Radiation response of
programmable logic devices depends on technology, fabrication process and design to be
implemented. Radiation testing is required on devices to verify that they are going to
meet the mission requirements. Radiation testing done on earth is very difficult to
perform because it is hard to predict the space environment, which depends on particle
types, energies and time of the mission.
The best way to test radiation hardness of semiconductor devices is to perform the
tests in space. This approach is very costly and is limited to technology demonstrations

55

and research type satellites. The only way to test programmable logic devices is through
ground based testing. The objectives of ground based testing are as follows:

Determine how the radiation effects a particular device and device material and what
failure rate.

Determine the response of the device from radiation effects based on technology and
the type of device.

Evaluate the performance of the device and verify if it has met mission requirements.
The radiation sources used for the testing are radioactive isotopes such as Co-60,

X-ray tubes and particle accelerators. Radiation tests are required to be performed at
worst case temperatures because it greatly affects the overall performance of the device.
Ground based radiation testing is done to check total dose effects, SEU and SEL,
transient response and neutrons.
4.5.6

Radiation Hard and Radiation Tolerant Devices

4.5.6.1

Actel

4.5.6.1.1

General Architecture

Actel has had the leading edge for RadHard and RadTolerant devices. The ACT
1, ACT 2, ACT 3 and SX architectures are available in RadHard or RadTolerant devices.
The logic block architecture, the routing resources, the I/O structure and the radiation
specifications will be discussed in the following sections.
4.5.6.1.1.1

ACT 1

ACT 1 architecture was the first done by Actel with a silicon structure and oxide
barrier. The ACT 1 devices consist of rows of logic blocks with 22 tracks of horizontal
routing separating the rows. The tracks are broken into segments that can be connected

56

together to have short wire segments or long wire segments. Short segments can be
connected by an antifuse to make long segments and there are programmable
interconnects between the horizontal and vertical tracks. The logic module in an ACT 1
device has 8 inputs and 1 output and can be configured into basic logic functions i.e.
AND gates and NOR gates or more complex logic functions, such as, latches and
multiplexors. ACT 1 contains a clocking network distributed throughout the chip to a
large number of flip-flops. Each logic module connects to horizontal tracks by vertical
connections to implement the functions for each logic module. The I/O ports can be
configured as input, output and bidirectional and would have to be sent through a logic
cell to get it registered. Figure 4.13 shows the ACT 1 logic block.

M1
D0
Y

M3

D1 S
D0

M2

D1 S

D0
Y
D1 S

Figure 4.13 ACT 1 Logic Block, 8 Input and 1 Output

57

4.5.6.1.1.2

ACT 2

ACT 2 FPGAs have two types of logic blocks. The two tyes are C-module for
combinatorial and S-module for sequential logic. The ACT 2 logic block consists of a 4
input multiplexor. This allows for wider fan in on logic functions. ACT 2 C-module is
shown in Figure 2.14. An S-module is a C-module driving a flip-flop with an additional
gate, which is the reset to the flip-flop. The row architecture consists of two C-modules
adjacent to two S-modules and so on. Act 2 devices also contain horizontal routing
channels that have segments that can be connected together by fuses. Vertical segments
on inputs and outputs connect the logic block to the horizontal channels using antifuses.
The input vertical segments on a logic module can be connected to the horizontal channel
above or below the logic module. The output vertical segments can span two rows each
direction. There are I/O modules at the end of each row and at the top and bottom of the
device. The I/O modules at he end of the row can accept I/O form the rows above and
below the module. The top and bottom I/O have a single source. ACT 2 supports tow
dedicated clock inputs or two internal clock inputs that are distributed down each
horizontal channel. ACT 2 has an improved internal layout allowing more routing
capability. For this reason the ACT family range in speeds form 65 to 130 MHz.
4.5.6.1.1.3

ACT 3

ACT 3 devices contain the same C and S modules as the ACT 2 devices. The
interconnection structure is the same as the ACT 2 devices except for additional vertical
track per pair of C and S modules. The I/O modules were enhanced. The I/O module
contains built-in flip-flops, dedicated clock and reset signals. Registered I/Os can be fed
back to the internal logic as used as registers for the design. ACT 3 contains a general

58

clock network and a high-speed hard-wired clock network (HCLK). This allows for
better speed performance with devices up to 150MHz clock frequencies with very low
skews.

D00
D01
Y

D10
D11
S1

S0

Figure 4.14 ACT 2 C-Module

4.5.6.1.1.4

Actel SX

The Actel SX is known for its sea-of-modules architecture instead of the rowbased architecture. The grid of logic modules allows for very little chip area lost to
routing resources. The SX family has two types of logic modules, the C-module and the
R-module. The R-module or the register module contains a flip-flop with more control
than previous Actel devices and architectures. It includes an asynchronous clear and
preset with clock enable lines. The R-module allows for programmable clock polarity on
a module by module basis. The clock for the R-module can be the hard-wired clock or
the routed clock. The C-module, or the combinatorial module, can be configured as a
wide range of logic function up to five inputs. The DB input and its inverter allow more
59

combinatorial logic functions than earlier Actel architectures increasing the options form
800 to 4000 logic functions. The Actel module organization in the SX devices is broken
into clusters. There are two types of clusters. A type-1 cluster contains two C-modules
and one R-module and the type-2 contains two R-modules and one C-module. The
clusters are grouped into two types of superclusters. Figure 4.15 identifies the C and Smodules and the supercluster configuration. Supercluster-1 contains two sets of type-1
clusters and supercluster-2 contains one type-1 and one type-2 cluster. SX devices
contain more supercluster-2 because most designs are more combinatorial in nature than
register intensive. Actel has provided flexible routing resources in the SX architecture
that reduces on the number of antifuses, which improves performance and allows them to
be fast and predictable. There are local routing resources called FastConnect and
DirectConnect.

DirectConnect is the hard-wired routing resources between the C-

module and the R-module within the clusters.

FastConnect is one programmable

connection between the logic modules within and cluster. Global routing is throughout
the chip providing interconnection within the superclusters and to the I/O pins. There are
three clocks in the SX chip and they are HCLK, CLKA and CLKB. CLKA and CLKB
are global clocks for internal logic and HCLK is a hard-wired clocking network for the
R-module.

60

C-Module

R-Module
DO

Routed
Data Input
S1
S0

D1

PSETB

D2
D

D3

Direct
Connect
Input

CLRB

HCLK

DB

CLKA,
CLKB
CKS

A0

CKP

Cluster 1

Sb

Sa

Cluster 1

B0

Cluster 2

Type 1 Supercluster

A1

B1

Cluster 1

Type 2 Supercluster

Figure 4.15 SX Module and Supercluster Architecture

4.5.6.1.2

Radiation Hard Actel Devices

4.5.6.1.2.1

RH1020 and RH1280

RH1020 and RH1280 have densities of 2,000 and 8,000 gates respectively and are
the ACT 1 architecture with 50 MHz on-chip performance. They are processed in 0.8
micron, tow-level metal bulk CMOS technology. They have row-based architecture and
Actels PLICE antifuse technology. These devices are the only RadHard high density
field programmable gate arrays on the market today with a total dose hardness of 300K
rad(Si), latch-up immunity and a single-event upset tested at less than 1x10-6 errors/bitday. Lockheed Martin Federal Systems in Manassas, Virginia has full Qualified
Manufacturers List (QML) certification on these devices.

QML indicates that the

performance and quality of the device is built into the device rather than tested at the end.
With the accumulated total dose effect at 300k rad (Si), these are ideal devices for deep
61

space spacecraft. They are ideal for high dose rate survivability from solar flares in earth
orbiting applications. Table 4.4 and Table 4.5 shows the profile of the devices and the
radiation specifications.
Device
Capacity

Logic Modules
Flip-Flops
Routing
Resources
And I/O

Gate Array Equivalent Gates

PLD Equivalent Gates
TTL Equivalent Packages
20-Pin PAL Equivalent Packages
Total Logic Modules
C-Modules
S-Modules
Maximum Flip-Flops
Horizontal Tracks/Channel
Vertical Tracks/Channel
PLICE Antifuse Elements
User I/Os

RH1020
2,000
6,000
50
20
547
547
0
273
22
13
186,000
69

RH1280
8,000
20,000
200
80
1,232
608
624
998
35
15
750,000
140

Table 4.4 RH1020 and RH1280 Device Profiles

Symbol Characteristics
RTD
Total Dose
SEL
Single Event Latch-Up
SEU1
SEU2
SEU3
RNF

Conditions

Min.

55C<T<125C

Single Event Upset for 55C<T<125C

S-modules
Single Event Upset for 55C<T<125C
C-modules
Single Event Fuse 55C<T<125C
Rupture
Neutron Fluence

Max.
300K
0
1E-6
1E-7
<1

>1E+12

Table 4.5 RH1020 and RH1280 Radiation Specifications

62

Units
Rad (Si)
Fails/Device
-Day
Upsets/BitDay
Upsets/BitDay
FIT
N/cm2

4.5.6.1.2.2

RAD-PAK

RAD-PAK is a packaging technology from Space Electronics Inc. (SEI) that

shield electronic components form total ionizing dose (TID) effects. This approach can
be used on non-radiation hardened devices used in applications that require hardened
devices. The RAD-PAK technology uses tungsten layer in the package lid to shield form
electrons. High atomic number materials are useful in scattering electrons. The RADPAK technology is used on Actel A1020 and A1280 devices and flown on some NASA
flight missions. RAD-PAK technology does not provide shielding form high-energy
protons or heavy ions that cause SEUs and latch-up. The RAD-PAK technology used on
Actel RP1280 and RP14100 introduce a higher density FPGA using ACT 1 and 2
architectures. This packaging technology eliminates the need for board and electronic
assembly level shielding. The RAD-PAK devices can be used in a wide subset of earth
orbiting and deep space missions. Table 4.6 shows the device profiles for Actels Rad
tolerant FPGAs with RAD-PAK packaging technology.
Device
Capacity

Logic Modules
System Speed
I/Os

Gate Array Equivalent Gates

PLD Equivalent Gates
TTL Equivalent Packages
20-Pin PAL Equivalent Packages
Total Logic Modules
C-Modules
S-Modules
Max. System Speed
User I/Os

RH1020
16,000
8,000
200
80
1,232
608
624
40 MHz
140

Table 4.6 Actels RAD-PAK Devices

63

RH1280
20,000
10,000
250
100
1,377
680
697
60 MHz
228

4.5.6.1.2.3

RadTolerant Devices

Actel has a wide range of devices that are RadTolerant. A device that is radiation
tolerant has not been designed or tested for radiation hardness but to some level of
radiation resistance.

RadTolerant devices are available in ACT 1,2,3 and SX

architecture. Radiation survivability for total dose results is summarized in two ways.
The first is the amount of total dose that is reached when the device fails to meet the
specification but still functions. The second is the total dose that is reached prior to
failure. The ability for the RadTolerant devices to survive radiation effect is dependent
upon the device and the lot. The 54SX family of devices have shown results form 60
100 Krad (Si) for not meeting the specification but still functioning and up to 240 Krad
(Si) for functional failure. Table 4.7 contains the 54SX Actel RadTolerant device profile
and Table 4.8 contains the radiation specifications.
Device
Gate Capacity
Logic Modules
Register Cells
Combinatorial Cells
Maximum User I/Os

RT54SX16
16,000
1,452
528
924
177

RT54SX32
32,000
2,880
1,080
1,800
224

Table 4.7 54SX Device Profile

Device

RT54SX16 REV 0
RT54SX16 REV 1
RH54SX16

Tech.

Bias
Voltage

Tolerance
Rad(Si)

Functional
Rad(Si)

SEU
LETth
MeVcm2/mg

Latch-up
LETth

0.6um
0.6um
0.6um

3.3/5.0
3.3/5.0
3.3/5.0

50-80K
100K
>200K

>240K

17
11
19

>80
>82
>80

Table 4.8 54SX Radiation Specification

64

ACT 1, 2 and 3 RadTolerant devices have the capability of reach gate capacities of
20,000, on-chip frequencies of 85 MHz and up to 228 User I/Os. The RT1020 devices
have results up to 100 Krads (Si) for ICC and >100 Krad (Si) for functional failure. The
ACT 2 devices have results up to 10 Krad for ICC and 18 Krad for functional failure.
The ACT 3 devices have typical results up to 28 Krad for ICC and 77 Krad for functional
failure. Table 4.9 contains Actels ACT 1, 2 and 3 RadTolerant device profile and Table
4.10 contains the radiation specifications.
Device
Capacity

Logic
Modules
Tech.
System
Speed
I/Os

Gate Array
Equivalent Gates
PLD Equivalent
Gates
TTL Equivalent
Packages
20-Pin PAL
Equivalent Packages
Total Logic
Modules
C-Modules
S-Modules
Actel Technology
Max. System Speed
User I/Os

RT1020
2,000

RT1280A
8,000

RT1425A
2,500

RT1460A
6,000

RT14100A
10,000

5,000

20,000

6,250

15,000

25,000

50

200

60

250

250

20

80

25

60

100

547

1,232

310

848

1,377

547
N/A
ACT 1
20 MHz

608
624
ACT 2
40 MHz

150
160
ACT 3
60 MHz

416
432
ACT 3
60 MHz

680
697
ACT 3
60 MHz

69

140

100

168

228

Table 4.9 ACT 1, 2 and 3 RadTolerant Device Profile

Device

RT1020
RT1280A
RT1425A
RT1460A
RT14100A

Tech.

Bias
Voltage

Tolerance
Rad(Si)
Max.

Functional
Rad(Si)
Max.

SEU Max.
LETth
MeVcm2/mg

Latch-up
LETth

1.2um
1.0um
0.8um
0.8um
0.8um

5.0
5.0
5.0
5.0
5.0

>100K
5-10K
20K
28K
20K

>100K
18K
40K
54-77K
>50K

25
26

>125
>120

32
30

>120
>120

Table 4.10 ACT 1, 2 and 3 Radiation Specification

65

4.5.6.2

Xilinx

4.5.6.2.1

General Architecture

As mentioned earlier, the first FPGAs introduced by Xilinx was in 1985 with the
XC2000 family.

Now there are four families of devices, which are the XC2000,

XC3000, XC4000 and XC5000. The configurable logic blocks (CLBs), the I/O blocks
and the routing resources of the XC4000 family will be discussed in this section. The
XC4000 is a very flexible family of devices and has recently introduced the XQR4000XL
radiation hardened FPGA for space and satellite applications. In general, Xilinx FPGAs
are a relatively course-grained structure that contains logic cells form 64 to 1024
corresponding to gate ranges from 6,000 to 25,000.
4.5.6.2.2

XC4000 Family

The Xilinx logic cell arrays (LCAs) are called configurable logic blocks (CLBs).
The CLB for the XC4000 family is more complex than the earlier families and are course
in structure than the Actel devices. The CLB consists of two four input logic blocks that
feed into a third logic block with a ninth input. A simple logic function or a combined
function can be derived form the two combinatorial outputs driven by the logic blocks.
The combinatorial outputs can feed the two flip-flops or they can be fed by an
independent set of signals. There are four outputs from the CLB, two are from the flipflops and two are from the combinatorial logic. Each flip-flop has a common control
line, programmable set and reset common enable, individual override and common clock.
The CLB architecture is ideal for fast arithmetic operations. For example each block can
be configured as a two-bit adder with fast carry. The excellent carry network can be used
as counters. 8 CLBs can be used to design a 16-bit up / down counter. The two large

66

function blocks inside a single CLB can be configured as a 32-bit RAM.

The

propagation delay and the access time are about 5.5 ns. Figure 4.16 shows a block
diagram of the Xilinx XC4000 configurable logic block.

C1

C2

C3

C4

H1 DIN S/R EC
S/R
CONTROL

G4
G3
G2

Logic
Function
of
G'
G1-G4

DIN
F'
G'
H'

G'
H'

EC RD

SD

Q2

G1

F4
F3
F2

Logic
Function
of
F1-44 F'

Logic
Function
H'
of
F',G' and
H1

F1

1
DIN
F'
G'
H'

S/R
CONTROL

K
(CLOCK)

Multiplexer
Controlled
By Configuration
Program

H'
F'

SD

Q1

EC RD
1

Figure 4.16 Xilinx XC4000 Configurable Logic Block (CLB)

The I/O blocks offer CMOS/TTL threshold selection, three-state control, slew rate
controls and pull-up and pull-down resistors. The XC4000 device has flip-flop / latches
for the I/O signals. The XC4000 family has a Wide Edge Decoder on the periphery of
the chip. Depending on the specific device they range from 42 to 72 inputs and can be
broken up into more narrow decoders. The advantage of this is to provide wide logic

67

functions just as the CPLDs. For example, it may take three to four levels of logic to
decode a bus instead of just one using the wide decoder architecture.
The cell interconnect structure of Xilinx devices are short and long vertical and
horizontal lines. The short lines make the interconnections between the switch matrices,
the CLBs and the long lines. The long lines carry signals throughout the devices with
high fan-out and low skew.

The XC4000 cell interconnect structure contains

symmetrical placement of the CLB I/Os, double length lines and a flexible simple switch
matrix. There are three types of interconnects and they are single-length lines, doublelength lines and long lines. The single length lines are the horizontal and vertical lines
that intersect at each switch matrix. There are programmable n-channel pass transistors
that connect the single-length lines in the switch matrix. The double-length lines are
metal segments that are twice the size of single-length lines and pass two CLBs before
they meet at a switch matrix. The long lines are metal interconnects that run the entire
width of the device. They are connected to the single-length lines via programmable
interconnect points. Long lines are used for high fan-out, low skew and time critical nets.
The XC4000 family can reach frequencies of 60 to 70 MHz because of the flexible
routing resources.
4.5.6.2.3

Radiation Hardened XC4000

Xilinx Corperation with Ericsson Saab Avionics conducted the first neutron
radiation tests on the XC4000-series devices for applications in an atmospheric radiation
environment.

The test concluded, that SRAM FPGAs can be used in atmospheric

radiation environment with results of one bit-error per 275,000 hours of operation at a
1000 km altitude at 60 degrees. Xilinx then proceeded to aim for demands for low earth

68

orbiting devices. The fabrication process was modified for the XQR series with 7 micron
epitaxial layer and implant layers to meet the CMOS threshold levels and performance.
The internal architecture and logic cells were not changed. Xilinx now offers radiation
hardened devices with gate capacities of 62,000. The Devices can handle a total dose of
60K rad (Si) and are immune to latchup at 100 MeV. They have incorporated two
methods to detect soft errors and hard errors caused by a single-event upset. These are
the ability to readback all the configuration data and provide triple redundancy error
detection. The readback function provides read-out capability of all the configuration
data while allowing the device to normally operate. The triple redundancy method
requires three FPGAs to detect an error, which will result in all the devices being
configured form the original source or the faulty device to be configured via readback
from the others. Table 4.11 and 4.12 contain the Xilinx radiation hardened device profile
and the radiation specifications respectively.

Device

Logic
Cells

XQR4013XL
XQR4036XL
XQR4062XL

1,368
3,078
5,472

Max.
Logic
gates
(No
RAM)
13,000
36,000
62,000

Max.
RAM
Bits
(No
Logic)
18,432
41,472
73,728

Typical Gate
Range (Logic and
RAM)

CLB
Matrix

Total
CLBs

# of
F-F

10,000-30,000
22,000-65,000
40,000-130,000

24x24
36x36
48x48

576
1,296
2,304

1,536
3,168
5,376

Table 4.11 Xilinx XC4000 Radiation Hardened Devices

69

User
I/O

192
288
384

Symbol
TID
SEL
SEU
SEU
SEU
SEU
SEU

Description
Total Ionizing Dose
Single Event Latch-up >100MeV cm2/mg +125C
Single Event Upset Galactic p+ 68km LEO
Single Event Upset Galactic Heavy Ion 68km LEO
Single Event Upset Trapped p+ 68km LEO
Single Event Upset Galactic p+ 35,000km GEO
Single Event Upset Galactic Heavy Ion 35,000km
GEO

Max.
60K
0
2.43E-8
9.54E-8
2.50E-7
5.62E-8
2.43E-7

Units
Rad (Si)

Table 4.12 Xilinx XC4000 Radiation Specification

4.6

Intellectual Property Cores for System-on-a-chip Applications

4.6.1

Intellectual Property Cores Introduction

Intellectual property (IP) cores are predesigned and preverifed logic functions for

ASIC and PLD-based designs. IP cores were originally provided to designers by third
party suppliers for ASIC system design and are now available for PLDs. PLD densities
have increased rapidly allowing for IP cores to be integrated into the design and still
allow plenty of room for interface logic. They have become high level building blocks
for complex designs.
IP cores come in three different forms i.e. hard cores, soft cores and the mixture
of the two, firm cores. Hard cores have been tested to have a proven layout and accurate
timing specifications.

Implementing a hard core into a PLD design is similar to

integrating a packaged semiconductor onto a board for discrete functions. Hard cores
have already been optimized and are ideal for timing critical applications. Hard cores
cannot be changed and are typically the property of large semiconductor companies like
Motorola and Texas Instruments.
Soft cores are more flexible than hard cores. A soft core is described at a
behavioral or Register Transfer Level (RTL) and in VHDL or Verilog. The soft cores are

70

designed to meet minimum specifications over a large range of device technology. Many
of the soft cores come with VHDL test benches. After synthesis the cores can be verified
with the test benches for post synthesis simulation. It is critical to modify the test bench
to include complete system level testing when the soft core is integrated with the
interface logic. Soft cores are more flexible and can be targeted to various PLD vendors
after post synthesis timing verification and simulation.
A firm core is a cross between a hard core and a soft core. A firm core is a
synthesized gate-level description of a function. A firm core cannot be changed but can
be simulated, placed and routed with the interface logic. Refer to Table 4.13 for the IP
cores, their representation and characteristics.

IP Category
Soft

Firm

Hard

Representation
Microprocessor targeted
C code
RTL HDL netlists

Characteristics
Flexible, technology-independent mapping
Slow system level simulation
Describes hardware architecture function
but not low level performance.
High IP security risk
Less flexibility

Assembly-code
algorithms
Gate-level HDL netlists
Generic technology
library
Timing models, floorplan
Object code / machine
code
Gate-level netlists

Very slow system-level simulation

Reliably predicts hardware low level
performance
High IP security risk
Least flexible, technology-specific fabprocess mapping
Very slow and/or difficult system-level
simulation
Predicts hardware performance most
reliably
Low IP security risk

Technology-specific,
process specific library
Timing models, routing
Polygon data

Table 4.13 IP Categories, Representation and Characteristics

71

System designers require the IP building blocks to be completely tested and

verified prior to integration into a complete design. IP testing is very complicated and
time consuming. The IP vendors are providing system-level evaluation models (SEMs)
to the designers. The SEMS allow the designers to verify the system level performance
of the design using algorithm/behavioral versions of the IP. Recently there has been two
standards put through the industry on IP packaging and interfaces. The first standard is
the Virtual Socket Interface (VSI). VSI standard outlines requirements for processor, bus
functional and behavioral IP cores and defines the set of documentation required by the
designers to integrate the IP blocks into their design. The Open Model Forum (OMF)
Standard defines a standard for IP core simulators. Open Model Interface (OMI) will
permit an OMF-compiled IP core to be simulated regardless of the developing language.
These standards will play an important role in the design process using IPs. The overall
design process can be broken down into three stages to make the design flow of complex
designs more manageable. The first stage is the design and verification of the system at
an individual block level. The second stage is to assemble the blocks into the system
design for virtual prototyping. The third stage is creating a physical prototype through
hardware emulation to reflect the functionality and the performance of the design in the
targeted device technology.
IPs can be used for system-on-a-chip designs that would include third party IP
processors, which can be integrated with peripheral devices, for example UARTs. There
are many PLD vendors that are providing IP cores.

72

4.6.2

Intellectual Property Core Vendors

IP cores are available from various FPGA vendors, which include Actel, Xilinx,

Altera, Lucent Technologies and Gatefield Corporation. Altera, Xilinx and Actel IP
cores will be discussed in the following sections.
4.6.2.1

Altera Megafunctions
Alteras Flex 10K family contains an embedded array of logic cells with large

blocks of RAM that can be integrated with other megafunctions.

The Flex 10K

architecture has the logic array blocks (LABs) with embedded array blocks (EABs). The
FLEX 10K architecture lends itself to IP core integration rather than for single
application designs.

Alteras MegaCore function and Altera Megafunction Partners

Program (AMPP) provide many different types of IP cores.

MegaCore functions are pre-optimized and tested logic blocks to perform a welldefined task in a system. The MegaCore functions can replace standard ICs or can
replace new functions, for example, processors and bus controllers. The MegaCore
functions come in VHDL, Verilog or Alteras Hardware Descriptive Language and can
be used in schematics, EDIF netlist files or HDL code. The MegaCore functions are
optimized for Alteras FLEX 8000 and FLEX 10K device families. The MegaCore
functions library contains the following eight megafunctions:

8237 DMA controller

6502 8-bit processor

6402, 16450, and 8251 UART processors

6850 Asynchronous communications interface adapter

8259 Interrupt controller

73

8255 parallel I/O controller

Create Initial Block
Diagram
Break the project into
functional blocks using topdown analysis

Identify Available
Megafunctions

Ship the Units

Identify functions using

reference materials.

Ahead of schedule.

Identify Megafunction
Source

Complete Design

Contact AMPP partners or

Altera.

Generate programming files

and test hardware.

Evaluate Specific
Megafunction

License Megafunction

Use the OpenCore evaluation

feature.

Contact Altera or AMPP

partner for terms and
conditions.

Install Megafunction

Simulate Design

Place in project hierarchy and

process as usual.

Change as necessary.

Figure 4.17 Altera Megafunction Top-Down Design Flow

The AMPP is an alliance between Altera and companies who specialize in
specific megafunctions to increase the use of IP cores in PLDs. This alliance focuses on
digital signal processing (DSP), bus interfaces such as PCI and USB and
communications. DSP megafuctions will support fast Fourier transforms (FFTs) and
adaptive impulse response (FIR) functions and have been provided by Silicon Systems
Ltd. Who specialize in DSP ASICs and DSP megafunctions. PCI megafunctions are

74

coming form Eureka Technology and running at speeds of 33 MHz. In the area of
communications AMPP joined with Nova engineering to provide parameterized complex
multiplier/mixer and quadrature output NCO.
An important factor with using IPs in a complex system is the verification and
validation that the core can meet the design requirements and specifications. Altera has
introduced the OpenCore evaluation system, which allows the designer to try the core and
see if it meets the specifications prior to licensing the megafunction. Figure 4.17 shows
Alteras Megafunction top-down design flow.
4.6.2.2

Actel Intellectual Property Cores

Actel is partnered with industry to provide IP cores that are optimized for the

specific architecture of Actel devices. The cores are avialable in hardware descriptive
language (VHDL or Verilog) or in netlist format for a specific device family or
technology. The documentation that is delivered with the core is the source code (HDL
or netlist), testbench, synthesis script, user guide and data sheets. All the cores are
designed, verified and tested in Actel FPGAs.

Table 4.14 contains the Actel core

application and description and the associated Actel target technology.

Application
Processor
Telecon

Interfaces

Description
8-bit microprocessor
All digital phase lock loop
ISDN E1 framer/deframer
High Level Data Link Controller
ATM Forum Utopia level II interface
Controller Area Network bus interface
I2C Master and Slave interfaces
Universal Asynchronous Receiver Transmitter
Interface (UART)
VME Slave Interface
Serial Communication Controller

Target Technology
MX,SX
MX,SX
MX,SX
MX,SX
MX
MX,SX
MX,SX
MX,SX
MX,SX
MX,SX

Table 4.14 Actel Core Applications and Description

75

4.6.2.3

Xilinx Intellectual Property Cores

The Xilinx XC4000 architecture is the key device for Xilinx CORE solutions.

Xilinx next generation FPGAs will have gate density ranges form 20,000 to 500,000 and
will soon be up to 1 million gates. The Xilinx Virtex family currently has 1500 to 32,000
logic cells, with on-chip frequencies of 100 MHz and an internal supply voltage of 2.5 V
and 3.3 V I/O buffers. The Xilinx CORE solutions contain predefined logic functions
that are optimized for Xilinx architecture to increase performance and assist in reducing
the development time of the design. The Xilinx CORE solutions are broken into four
groups supporting four applications areas, which are as follows:

Xilinx CORE solutions applications

1. Standard Bus Interfaces such as PCI, PCMCIA, USB and IIC.
2. DSP Functions These range from small functions to large functions. The
small functions include adders, registers and multipliers. The large functions
include FIR filters and transforms.
3. Communications and Networking These functions include Asynchronous
Transfer Mode (ATM) cores, Forward Error Detection cores and
Telecommunication cores, such as HDLC Protocol and MT1F T1 Framer.
4. Base-Level Functions - These include small cores, such as, decoders,
multiplexers, integrators and parallel multipliers and large cores like UARTs
and RISC processors.

76

Xilinx CORE solutions groups

1. LogiCORE Products These are intellectual property cores that have been
pre-implemented and verified logic functions sold and supported by Xilinx.
2. AllianceCORE Program This is the alliance between Xilinx and third-part
IP vendors. This provides many more available IP products.
3. Reference Designs Reference designs allow designers to practice with IP
cores in system level architecture and are free of charge.
4. LogiBLOX Products Graphical tool for creating simple modules and using
generic modules in a logic implementation. LogiBLOX comes as part of the
Xilinx standard set of software.
Through research the Xilinx CORE solutions was the largest set of Intellectual

Property Cores available for FPGAs. The IP cores of interest for the implemented design
covered by this thesis are the UART and certain DSP functions. The current design
contains an external UART that could just as well been implemented using a Xilinx
XC4000 device and an IP core. It is common practice for camera systems to have image
preprocessing capability prior to data analysis. Common filters and transforms in DSP
cores can easily be implemented into the system design for distribution of device
overhead and data processing and is being considered for the next generation camera
interface electronics. The detailed design of the APS camera system and the interface
electronics is contained in chapters that follow.
4.7

Programmable Logic Devices Summary

Programmable Logic Devices (PLDs) come in two general categories.

The

categories are Simple Programmable Logic Devices (SPLDs) and High-Density

77

Programmable Logic Devices (HDPLDs). HDPLDs come in two architectures, which

are Complex Programmable Logic Devices (CPLDs) and Field Programmable Gate
Arrays (FPGAs). Both CPLDs and FPGAs are available in SRAM based programming
configuration but only CPLDs can be EPROM or EEPROM programmed. This means
that CPLDs can be up and running when power is applied and are nonvolatile.

Some

FPGAs are antifuse-based allowing them to only be one time programmable devices
(OTPs) and SRAM based, which can be programmed as many times as required. The two
major elements of CPLDs and FPGAs are the logic elements and the interconnect
structure. Both FPGAs and CPLDs have their advantages and disadvantages. CPLDS
are ideal for high-speed applications requiring critical timing and FPGAs are more
flexible with the finer-grained architecture. Lattice semiconductor CPLDs were used for
the design, specifically the ispLSI 3000 series.

This CPLD architecture offered

predictable timing, high densities, in-system programmability, flexible architecture for

mixed combinatorial and register intensive designs and system partitioning.

Some

applications can not use CPLDs. Planetary Spacecraft and earth orbiting satellites and
science instruments require Radiation Hardened PLDs. The mission length and orbit will
determine the levels of radiation the device will have to withstand. Actel is the leading
FPGA vendor for RadTolerant and RadHard devices with Xilinx just recently releasing
its first RadHard family of devices, the XC4000 series. Intellectual property (IP) cores
are the latest trend in the FPGA market. IP cores are predesigned and preverifed logic
functions for ASIC and PLD-based designs. They range in complexity form larger
processor and DSP cores to smaller multiplexer and UART cores. Three companies were
outlined, and those were Actel, Xilinx and Altrea.

78