Xcell Journal Online
  Xcell Journal Archives
   
  Writing for Xcell
  Advertising in Xcell
  FREE Subscription
   
  Partner Yellow Pages
  Reference Pages
  Contact Us

    

Home : Documentation : Xcell Journal Online : Article
Achieve Faster Timing Closure with Graph-Based Physical Synthesis



by Jeff Garrison, Director of Marketing, FPGA Products, Synplicity, Inc.
jeff@synplicity.com (12/1/05)


Graph-based physical synthesis was invented to improve timing closure by means of a single-pass physical synthesis flow.
article link to PDF
Article PDF 270 KB


Advances in FPGA technology have opened the door wide open for use in all types of applications, including wireless communications, computer, industrial, defense/aerospace, medical, automotive, and even consumer. Xilinx® Virtex™-4 devices have the capacity, performance, and cost structure to lead a migration from traditional cell-based ASICs to programmable devices in all but the highest volume and bleeding-edge applications. Along with this capability, however, are new challenges from a designer’s perspective. In this article, I’ll discuss a solution to one of these most important challenges – timing closure.

One of the biggest reasons to use FPGAs in the first place is their ability to deliver working silicon to an electronics system quickly and reliably. As the complexity of FPGAs and all integrated circuits has increased, the time-to-market advantage offered by FPGAs could be diminished if you do not make significant changes to your core design technology. The primary issue is timing closure – the ability to reach your design’s timing goals in a fast, predictable way. Gone are the days when logic delay and wire-load models for interconnect delay are enough to estimate timing and give predictable results. In 90 nm FPGAs, you must incorporate actual routing delay into the synthesis process to achieve rapid timing closure for high-performance designs.

Timing Accuracy is Everything
The underlying problem that determines if you will be able to close on timing is estimation accuracy. Historically, synthesis and placement tools have been based on the assumption that the proximity of logic and wire-load estimation determines the routing delay. Although this used to work reasonably well for ASIC design, it does not work at all for FPGAs. Unlike an ASIC, FPGA routing is pre-determined. In an ASIC the routing is customized for the placement of the logic. In other words, once the placement of an ASIC is done, it is relatively easy to get a good estimation of routing delay by measuring Manhattan distances from one point to another (see Figure 1).

Because FPGAs have fixed routing resources, the design tool needs to understand the different types of routing and its implication on timing. In an FPGA, the fastest routing between two points may very well not be the shortest. Think of your commute to work – sometimes it’s faster to go slightly out of your way and get on a freeway than to travel the shorter distance on side streets. The same concept applies to FPGAs: some direct routing resources (freeways) are faster than those that have to go through switch matrices (side streets) (Figure 2).

Figure 3 illustrates how placement is different for proximity- and graph-based tools.

Graph-Based Physical Synthesis
Synplicity invented graph-based physical synthesis to improve timing closure by means of a single-pass physical synthesis flow for 90 nm FPGAs. The essence of the graph-based approach is that the pre-existing wires, switches, and placement sites used for routing an FPGA can be represented as a detailed routing resource graph. The notion of distance then changes from proximity to a measure of delay and wire availability.

Synplicity’s graph-based physical synthesis technology merges optimization, placement, and routing to generate a fully placed and physically optimized design, providing rapid timing closure and a 5% to 20% timing improvement.

Graph-based physical synthesis does not require you to create a floorplan or provide process (often only known by expert users) in order to get good results. It is a fully automated methodology that can be used without special knowledge of the physical FPGA device. In addition to this fully automated mode, you do have the option to guide physical synthesis by providing design planning information (such as a floorplan) used during the physical synthesis process.

Synplify Premier
To directly address the challenge of keeping timing closure under control for advanced FPGA technologies, Synplicity has introduced Synplify Premier, its first FPGA design product based on graph-based physical synthesis technology. Synplify Premier includes all of the features in Synplify Pro and adds graph-based physical synthesis for the Virtex-4, Virtex-II Pro, and Spartan™-3 families. In addition to the new graph-based physical synthesis, Synplify Premier also offers a new capability for debugging and prototyping ASICs using FPGAs. One such technology is RTL instrumentation and debugging of live, running FPGAs.

This technology is based on Synplicity’s Identify product, which allows you to navigate your design graphically and mark signals directly in your RTL code as probes or sample triggers. After synthesis, you can view the signal values of a live, running FPGA directly in the RTL source code or in waveform. An incremental place and route capability saves time by allowing you to quickly update instrumented nodes and debug. The debugging technology within Synplify Premier software is closely integrated with synthesis and Xilinx ISE™ software for a seamless development environment.

A second technology important for ASIC prototyping with FPGAs is the ability to convert gated clocks to FPGA clock-enable structures without modifying your RTL source. Synplify Premier performs this task automatically, along with handling generated clocks and instantiations of most common Synopsys DesignWare components.

Conclusion
Because of the increased design complexity enabled by new devices such as Virtex-4 FPGAs, designers need EDA tools that can handle the physical properties of FPGA architecture to achieve acceptable timing closure. Several physical design tools based on ASIC technologies have been used to address FPGA design, but they have had little success. The ASIC approach does not work for FPGAs because the silicon fabric is completely different and, unlike ASICs, proximity does not imply better timing.

Synplicity’s Synplify Premier product with graph-based physical synthesis directly addresses the challenges of FPGA physical design and results in faster designs done in less time. For more information on graph-based physical synthesis, ASIC prototyping, and Synplify Premier, visit www.synplicity.com/products/index.html.

Printable PDF version of this article with graphics. PDF logo (12/1/05) 270 KB

 
Jobs Events Webcasts News Investors Feedback Legal Privacy Trademarks Sitemap
© 1994-2008 Xilinx, Inc. All Rights Reserved.