FPGA and ASIC Technology Comparison Part 1. Fundamentals of FPGA Design 1 day Designing for...
-
Upload
corbin-sarver -
Category
Documents
-
view
223 -
download
1
Transcript of FPGA and ASIC Technology Comparison Part 1. Fundamentals of FPGA Design 1 day Designing for...
FPGA and ASIC Technology Comparison
Part 1
Fundamentals of FPGA Design
1day
Designing forPerformance
2days
Advanced FPGAImplementation
2days
Intro to VHDL or Intro to Verilog
3days
FPGA and ASIC Technology Comparison
FPGA vs. ASIC Design FlowASIC to FPGA
Coding Conversion
Virtex-5 Coding Techniques Spartan-3 Coding Techniques
Curriculum Path
forASIC Design
Minimum
: 6 months design
experience
FPGA and ASIC Technology Comparison
Welcome
If you are an experienced ASIC designer transitioning to FPGAs, this course will help you reduce your learning curve by leveraging your ASIC experience
Careful attention to how FPGAs are different than ASICs will help you create a fast and reliable FPGA design
Objectives
After completing this training you will be able to:
Describe the differences between ASIC and FPGA architectures
Explain the features of Xilinx FPGA architecture
Benefit from the Xilinx dedicated resources
Contrasting Architectures
ASIC architecture compared to the Xilinx FPGA architecture– Gates versus LUTs– Delays– Performance
Fundamental part selection considerations– Cost– Size– Performance– Volume– Analog circuitry– Time to market– Reprogrammability
Standard Cell
Advantages– Lowest price for high-volume
production (greater than 200K per year)– Fastest clock frequency (performance)– Unlimited size– Integrated analog functions
• Custom ASICs– Low power
Disadvantages– Highest non-recurring engineering
costs– Longest design cycle– Limited vendor IP with high cost– High cost for engineering change
orders
Embedded Array
Advantages– Low price for medium-volume to
high-volume production– Performance only slightly slower than a
standard cell– 50+ million gates– Custom macro support– More flexibility than an FPGA– Low power
Disadvantages– High non-recurring engineering costs– Design cycle longer than an FPGA – Vendor IP has high cost– Generally digital only
Xilinx FPGAsField-Programmable Gate Arrays
Advantages– Lowest cost for low-volume to medium-
volume production– No non-recurring engineering costs– Standard product – Fastest time to market– Xilinx has extensive library of IP
• Inexpensive compared to ASIC vendors– Ability to make bug fixes quickly
and inexpensively
Disadvantages– Slower performance– Size limited to ~25 million system gates– Digital only
Field-Programmable Gate Arrays
Xilinx FPGAs are made using SRAM
Today’s FPGAs use 65-nm copper CMOS process
Potential to accommodate 25M system gates – Includes RAM and logic gates
Performance up to 550 MHz
Integrated synthesis, simulation, and place & route tools– PC and UNIX– Inexpensive: $2500 or less for the ISE Design Suite
• Use of third-party tools will increase costs• Free ISE WebPACK is available
Configuration Introduction
When does configuration happen?– On power up– On demand
Why do FPGAs need to be configured?
− FPGA configuration memory is volatile
− Configuration data is stored in a PROM or other external data source
What do you need to know about FPGA configuration?− What happens during configuration
− How to set up various configuration modes and daisy chains
Configuration
Cost of ownership is reduced with the ability to reconfigure the hardware—extending the life of the product
• Reduces the costly physical deployment of repair technicians
• Extends the life of the product Upgrades Bug fixes Adding additional functionality Faster time to market Partial reconfiguration
FPGAFPGA
FPGA Configuration Methods
Xilinx Cables:JTAGSlave SerialSlave SelectMAP
Microprocessor:JTAGSlave SerialSlave SelectMAP
Xilinx PROMs: Slave/Master Serial Slave/Master SelectMAP
Commodity Flash:Slave SelectMAPSPI*BPI*
*SPI and BPI support is available in the newer Virtex™-5 and Spartan™-3E families
Five Primary Elements
Routing
Xilinx FPGAs
Dedicated blocks
Input and output blocks
Configurable logic blocks
* Clocking Resources
Logic Cells
Logic cells include– Combinatorial logic, arithmetic
logic, and a register
Combinatorial logic is implemented using Look-Up Tables (LUTs)
Register functions can include latches, JK, SR, D, and T-type flip-flops
Arithmetic logic is a dedicated carry chain for implementing fast arithmetic operations
CarryChain
LUTCarry in
Carry out
D Q
S/R
Combinatorial Logic
LUT
LUTs function as a ROM
Combinatorial Logic
Z
They generate the output value… for a given set of inputs
AB
CD
FE
0 0 0 0 0 0 00 0 0 0 0 1 00 0 0 0 1 0 00 0 0 0 1 1 10 0 0 1 0 0 1
0 0 0 1 0 1 1
. . .
0 0 1 1 0 0 00 0 1 1 0 1 00 0 1 1 1 0 00 0 1 1 1 1 1
A B C D E F Z
0 0 0 1 0 1
Constant delay through a LUT
Limited by the number of inputs and outputs, not by complexity
Wide Input Functions
For wider input functions, LUTs can be combined using a multiplexer
These muxes are dedicated, so they are fast
LUT
LUT
LUTMUX
LUT-Based Memory
Can store 64 bits of memory as either a RAM or a ROM
Fundamentally, the LUT is a ROM
Can become RAM with activation of configuration write strobe
Combine multiple LUTs for larger memories—larger in both in depth and width
128 x 8 is not uncommon
6-input LUT contains two 5-input LUTs, which adds more flexibility
LUT
Carry Logic
The carry logic chain is dedicated logic that computes high-speed arithmetic logic functions
The carry chain generally consists of a multiplexer and an XOR gate– The LUT computes the multiplexer selector – The multiplexer determines the carry-out– The XOR gate computes the addition
Memory Blocks
Support single- and dual-port synchronous operations In dual-port mode, these RAM blocks support fully independent ports for both reading and writingEach RAM block can be configured for 36 kb– Can be used as 2 independent 18-kb
RAMs
Dedicated cascade logic allows 2 RAM blocks to be configured as 72 kbBlocks of memory are generally spread out across the dieDedicated FIFO logic enables each RAM to be configured as a FIFO
Port A: 8 bits
Port B: 32 bits
Configuration Depth Data Bits Parity Bits
32k x 1 32 kb 1 0
16k x 2 16 kb 2 0
8k x 4 8 kb 4 0
4k x 9 4 kb 8 1
2k x 18 2 kb 16 2
1k x 36 1 kb 32 4
Block RAM Configurations
Configurations available on each port
Independent configurations on ports A and B, read and write– Supports data-width
conversion, including parity bits
IN 8 bitIN 8 bit
OUT 32 bitOUT 32 bit
IOB Element
Input path– Two DDR registers
Output path– Two DDR registers– Two 3-state enable DDR registers
Each path can be combinatorial or registeredSeparate clocks and clock enables for I and OSet and reset signals are shared
IOB Element
Default I/O standard varies by family– Fast and slow slew rate– Programmable drive strength– Other I/O standards
Built in SERDES functionality– ISERDES divides input data by up to 10– OSERDES multiplies output data by up to 10
DSP Slice
25x18 Multiply ALU Mode
Pattern DetectionIndependent
C input
Dedicated ACascading
Routing
A combination of programmable and dedicated routing lines
Dedicated routing– Global clocks with predefined clock tree– Regional clocks and IO clocks– Global low-skew routing resources for other
high-fanout signals– Carry chain routing– Dedicated routing among other dedicated
resources
General interconnect– Routing of local signals between CLBs and
IOBs
Clock Management
Dedicated clock trees are pre-optimized clock networks that balance the skew and minimize delay
Virtex-5 FPGA has 32 separate clock networks
Spartan-3 FPGA has 8 separate clock networks
Each can be configured for a built-in clock enable (BUFGCE) or switching clock sources (BUFGMUX)
Local clock routing includes regional (BUFR) and SERDES (BUFIO)
Clock Management
PLL
Digital Clock Manager (DCM) consists of…
– Digital Delay Locked Loop (DLL)
– Digital Frequency Synthesis (DFS)
– Digital Phase Shifter (DPS)
CMT
I/O Translators
Programmable input and output thresholds
Supported standards include– LVCMOS (several classes), LVPECL, HSTL
(several classes), SSTL (several classes), PCI,
PCI-X, LVDS (several classes), GTL, GTL+, and
HyperTransport™ (LDT) technology
- Supported standards vary, check your data sheet
Different I/O standards require a separate input and output reference voltage for each bank supporting a separate I/O standard
Generally, each bank can support several standards, as long as they share the same vref (input) or vcco (output)
Dedicated and Special ResourcesClock management (CMT)– DCM and PLL– Dedicated clock trees (not
shown)
Test logic– Built-in JTAG
I/O translators– Supporting many different thresholds
Other resources– Dual-Data Rate (DDR) registers in IOB
– SERDES resources
Dedicated Cores– Block RAM– DSP Slices– Gigabit transceivers, MGTs (all devices)– Tri-mode Ethernet MAC (all devices)– PCI Express® core (all devices)
Additional FXT Cores– PowerPC® 440 processors (not shown)– Faster GTX transceiver (not shown)
Other Resources
Embedded processor cores– 32-bit PowerPC 440 processor core (hard)– MicroBlaze processor core (soft)
Digitally controlled termination resistance (DCI)
FPGA flexibility– Reconfigurable logic– Time to market– Lowest “cost of change”
Xilinx combinatorial resources use flexible LUTs
Xilinx slices also contain registers, carry logic, clocking resources, and dedicated muxes to improve the performance for all applications
Xilinx FPGAs have dedicated resources for DSP, RAM, PCI, EMAC, and I/O that make these critical paths equivalent to a custom ASIC
Summary
Where Can I Learn More?
Xilinx online documents – www.support.xilinx.com
• Software manuals• Data sheets• Application notes• User guides
Xilinx Education Services courses– www.xilinx.com/training
• Xilinx tools and architecture courses• Hardware description language courses• Free Videos
FPGA and ASIC Technology Comparison
Part 2
Fundamentals of FPGA Design
1day
Designing forPerformance
2days
Advanced FPGAImplementation
2days
Intro to VHDL or Intro to Verilog
3days
FPGA and ASIC Technology Comparison
FPGA vs. ASIC Design FlowASIC to FPGA
Coding Conversion
Virtex-5 Coding Techniques Spartan-3 Coding Techniques
Curriculum Path
forASIC Design
FPGA and ASIC Technology Comparison
Welcome
If you are an experienced ASIC designer transitioning to FPGAs, this course will help you reduce your learning curve by leveraging your ASIC experience
Careful attention to how FPGAs are different than ASICs will help you create a fast and reliable FPGA design
Objectives
After completing this training you will be able to:
Describe how a simple logic implementation can differ between ASIC and FPGAs
Recognize gate counts as an estimation of design size
Explain some of the FPGA design practices you must follow to get peak performance in your FPGA
Gate Comparison
In retargeting HDL code for an ASIC design to an FPGA, gate conversion is rarely one to one
A 0.13-µ standard cell can have up to 100K gates per mm2
– A Virtex®-5 FPGA has about 20K usable gates per mm2
Why the difference?
Xilinx has programmable logic in addition to the functional logic– Routing– Multiplexers– Configuration memory registers
This means built-in design flexibility!
Gate Translation
Separate out logic, flip-flops, RAM, cores, and I/O– Partition cores into logic and RAM
Assume– 6 to 24 gates per LUT (depending on the number of inputs used)– RAM bits are equivalent– Up to 100 ASIC gates per I/O; translate to IOBs– 7 gates per register
So what design strategy do you think you need to use?– To get the most out of the FPGA try to use as many features as possible,
especially the FPGA’s dedicated hardware
Example
ASIC 250K logic gates
Four 32-kb blocks of RAM
243 pads, includingpower and ground
FPGA 20,800 to 41,600 LUTs
Equivalent
Equivalent number of pins
Depending on the number of LUTs needed, this design could use a
Virtex-5 LX30, LX50, or LX85 FPGA
Any ASIC-to-FPGA gate counting method is only a rough estimate.
Taking ASIC code directly to an FPGA will not utilize the dedicated resources of the FPGA.
Gate Counts
Gate counts are influenced byCoding styleMetal layersProcess geometryLibrary qualityPlacement and routing algorithmsCore contents (RAM versus gates)I/O requirementsSpecial features
CONCLUSION
8-input AND gate
AND Gate Example
For vec(7.0)
assign and_out = & vec;
Verilo
gV
erilog
For vec(7 downto 0)
and_out <= vec(0) AND vec(1) AND vec(2) AND vec(3) AND vec(4) AND vec(5) AND vec(6) AND vec(7);
VH
DL
VH
DL
ASIC Implementation
8-input AND gateTwo four-input NAND gates feeding a two-input NOR gate
Approximate delay in a standard-cell ASIC with 0.13-µ process = 0.47 ns
Approximate gate count = 14
Beware of ASIC libraries with very wide gate types!
Xilinx Implementation
Approximate max delay in a Virtex-5 FPGA = 0.435 ns
Approximate gate count = 18 gates
8-input AND gate implemented in three 4-input LUTs and two logic levels
Approximate max delay in a Spartan®-3 FPGA = 0.678 ns
Approximate gate count = 18 gates
Question
How many 4-input LUTs would be required to implement a 32-input OR gate?
How many Logic Levels would they generate?
Answer
How many 4-input LUTs would be required to implement a 32-input OR gate? 11
How many Logic Levels would they generate? 3
If net delays ~ .3 ns and LUT delays ~.2 ns then total delay would be 2(.3) + 3(.2) ~ 1.2 ns
…in a Spartan®-3 FPGA
How do you think this would be implemented in Virtex-5 with a 6-input LUT? (Answer: 7 LUTs and 2 Logic Levels)
LUTLUT
LUTLUT
LUTLUT
LUTLUT
LUTLUT
LUTLUT
LUTLUT
LUTLUT
LUTLUT
LUTLUT
LUTLUT
Tri-State Busses
Some ASIC designs have large tri-state busses– There are no tri-state buffers associated with each slice in the newest
FPGAs– These will have to be re-synthesized and be mapped to LUTs and the F7
and F8 dedicated muxes– You may need to code these with a CASE statement and a high-Z output– The F7 can implement an 8-to-1 mux – The F8 can implement a 16-to-1 mux
Registered AND gate
process (clk)beginif rising_edge(clk) then
vec_q <= vec;and_out <= vec_q(0) AND vec_q(1) AND vec_q(2) AND vec_q(3) AND vec_q(4) AND vec_q(5) AND vec_q(6) AND vec_q(7);
end if;end process;
VH
DL
VH
DL
always @ (posedge clk)begin
vec_q <= vec;and_out <= & vec_q;
end
Verilo
g
Performance Comparison
A comparison of the achieved performance for the registered 8-input AND gate– Virtex-5 FPGA
• ~550 MHz• ~88 gates
– 0.13-µ standard cell ASIC• ~850 MHz• ~77 gates
Typical high-performance frequencies (no optimization for the FPGA)– Virtex-5 FPGA
• ~275 MHz for four-levels of LUT (combinatorial) logic– 0.13-µ standard cell ASIC
• ~550 MHz for equivalent logic
Don’t forget to optimize your HDL code!
ASIC versus FPGA
Combinatorial logic implemented in an ASIC is typically faster than in an FPGA implementation– The fine-grain architecture of an ASIC allows wider input functions
to be implemented with significantly less delay– ASICs have a dedicated routing structure rather than a
programmable routing structure
Critical paths typically include I/O, RAM, PCI™ technology, EMAC, and DSP resources– Xilinx has dedicated FPGA resources to implement these
functions, making these paths equivalent to an ASIC implementation• Remember: Xilinx Virtex-5 devices are cutting-edge ASICs
Don’t forget to include Xilinx-dedicated resources in your design!
Pipelining
fMAX = n MHz
D Q Two Logic Levels D Q
fMAX 2n MHz
One Level
One Level
D Q D Q D Q
Sequential Design
How do you get high performance from an FPGA?
Pipelining– For large combinatorial paths, additional registers may need to
be inferred to break up combinatorial paths to increase performance
– This technique increases the size of the design– This is not as likely to be needed for Virtex-5 FPGA designs
because the Virtex-5 FPGA has a 6-input LUT– Evaluate the number of logic levels your design has by
generating a timing report from the ISE® Design Suite or your synthesis tool
– Usually the registers are added at a hierarchical boundary
Don’t forget to evaluate the number of logic levels for your timing-critical paths!
Timing Constraints
How do you get high performance from an FPGA?
Timing constraints– Timing constraints communicate the performance goals to the
implementation tools– Global timing constraints constrain virtually all the paths in your design
based on your system frequency, input, and output times (PERIOD, OFFSET IN, OFFSET OUT)
– Path-specific timing constraints need to be added to constrain multi-cycle paths and false paths
Adding timing constraints is essential if you want good system speed!
Coding Style
How do you get high performance out of an FPGA?
Coding style has a large impact on the performance– Because FPGA combinatorial and routing resources are inherently slower,
the HDL coding style needs to be improved – Write your code to limit the number of logic levels inferred– Learn about proper HDL coding styles by listening to the REL modules
Don’t waste time! Evaluate your HDL!
Synchronous Design
How do you get reliability out of an FPGA?
Always build a synchronous design– Asynchronous circuits are less reliable– Lot variations exist for all FPGAs, which means that your design has to be
able to work for faster devices
Timing constraints– Cannot fix asynchronous design problems—only you can
Synchronous Design Methodology
One clock (or at least as few as possible)
Use one edge (all flip-flops use rising or falling edge)
Use D-type flip-flops
Register the outputs of each behavioral block
In place of multiple clocks, use clock enables
Synchronize asynchronous signals to the “single” clock (synchronization circuits)
Do NOT create– Gated, derived, or divided clocks– Local asynchronous set/reset– Avoid global asynchronous set/reset
Get it right the first time!
Summary
Don’t worry too much about gate counting methodologies. They are only rough estimates, anyway
Optimize your HDL coding style
Instantiate Xilinx-dedicated hardware resources into your design to improve your system speed and maximize what you get from your FPGA
Pipeline your timing-critical paths
Timing constraints are a primary means for improving system speed
Get your design to work properly the first time by designing synchronously
Where Can I Learn More?
Xilinx Answers Browser – www.support.xilinx.com Answers Browser window
• Enter keywords like “pipelining” or “period constraint”
Xilinx Education Services courses– www.xilinx.com/training
• Xilinx tools and architecture courses Fundamentals of FPGA Design
» Learn about synchronous design, global timing constraints, the Architecture Wizard, and the CORE Generator™ tool
Designing for Performance» Learn about avoiding metastability, path-specific timing
constraints, and the Timing Analyzer • Free Video-based Training
» Learn about proper HDL coding techniques
Xilinx is disclosing this Document and Intellectual Property (hereinafter “the Design”) to you for use in the development of designs to operate on, or interface with Xilinx FPGAs. Except as stated herein, none of the Design may be copied, reproduced, distributed, republished, downloaded, displayed, posted, or transmitted in any form or by any means including, but not limited to, electronic, mechanical, photocopying, recording, or otherwise, without the prior written consent of Xilinx. Any unauthorized use of the Design may violate copyright laws, trademark laws, the laws of privacy and publicity, and communications regulations and statutes.
Xilinx does not assume any liability arising out of the application or use of the Design; nor does Xilinx convey any license under its patents, copyrights, or any rights of others. You are responsible for obtaining any rights you may require for your use or implementation of the Design. Xilinx reserves the right to make changes, at any time, to the Design as deemed desirable in the sole discretion of Xilinx. Xilinx assumes no obligation to correct any errors contained herein or to advise you of any correction if such be made. Xilinx will not assume any liability for the accuracy or correctness of any engineering or technical support or assistance provided to you in connection with the Design.
THE DESIGN IS PROVIDED “AS IS" WITH ALL FAULTS, AND THE ENTIRE RISK AS TO ITS FUNCTION AND IMPLEMENTATION IS WITH YOU. YOU ACKNOWLEDGE AND AGREE THAT YOU HAVE NOT RELIED ON ANY ORAL OR WRITTEN INFORMATION OR ADVICE, WHETHER GIVEN BY XILINX, OR ITS AGENTS OR EMPLOYEES. XILINX MAKES NO OTHER WARRANTIES, WHETHER EXPRESS, IMPLIED, OR STATUTORY, REGARDING THE DESIGN, INCLUDING ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE, AND NONINFRINGEMENT OF THIRD-PARTY RIGHTS.
IN NO EVENT WILL XILINX BE LIABLE FOR ANY CONSEQUENTIAL, INDIRECT, EXEMPLARY, SPECIAL, OR INCIDENTAL DAMAGES, INCLUDING ANY LOST DATA AND LOST PROFITS, ARISING FROM OR RELATING TO YOUR USE OF THE DESIGN, EVEN IF YOU HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. THE TOTAL CUMULATIVE LIABILITY OF XILINX IN CONNECTION WITH YOUR USE OF THE DESIGN, WHETHER IN CONTRACT OR TORT OR OTHERWISE, WILL IN NO EVENT EXCEED THE AMOUNT OF FEES PAID BY YOU TO XILINX HEREUNDER FOR USE OF THE DESIGN. YOU ACKNOWLEDGE THAT THE FEES, IF ANY, REFLECT THE ALLOCATION OF RISK SET FORTH IN THIS AGREEMENT AND THAT XILINX WOULD NOT MAKE AVAILABLE THE DESIGN TO YOU WITHOUT THESE LIMITATIONS OF LIABILITY.
The Design is not designed or intended for use in the development of on-line control equipment in hazardous environments requiring fail-safe controls, such as in the operation of nuclear facilities, aircraft navigation or communications systems, air traffic control, life support, or weapons systems (“High-Risk Applications”). Xilinx specifically disclaims any express or implied warranties of fitness for such High-Risk Applications. You represent that use of the Design in such High-Risk Applications is fully at your risk.
© 2012 Xilinx, Inc. All rights reserved. XILINX, the Xilinx logo, and other designated brands included herein are trademarks of Xilinx, Inc. PowerPC is a trademark of IBM, Inc.
Trademark Information