FPGA targeted design platforms: Fulfilling the programmable imperative
Moshe Gavrielov, who is president and CEO, Xilinx, discusses the company’ plans to bring customers Targeted Design Platforms, described as very defined and refined flows closely tied to Xilinx silicon.
For the past 25 years FPGAs have relentlessly increased in logic cell counts, functionality, and lower cost per transistor. In doing so, FPGAs have steadily taken market share away from the gate array and ASIC markets. Roughly 15 years ago, this FPGA progression helped render gate arrays obsolete. Today, numerous trends – the exorbitant cost of designing and manufacturing ASICs, ever-changing and not-yet-defined standards, the need to reduce Bill of Materials, and the need for both hardware and software programmability, all in the face of rough economic times and reduced staffing – are converging to create an environment where electronics products designers are, at a greater pace, dumping ASICs in favor of FPGAs. I call the convergence of these trends the programmable imperative.
FPGAs are available today with several hundreds of thousands of programmable logic cells, up to 11.2 Gbps transceivers, 38 Mb of BlockRAM, and 2,000 DSP slices. Designers are leveraging these FPGAs to address an ever-growing number of applications. All things considered, the opportunity is ripe for FPGAs to step up the pace in gobbling up more ASIC market share and claim a more central role as the heart of next-generation electronic innovations.
Xilinx is actively moving to seize the opportunity and help customers get their FPGA-based innovations to market quickly. With this goal in mind as we introduced our Virtex-6 and Spartan-6 FPGA families this year, the need for hardware- and software-design tools, IP, development boards, and support was emphasized. Xilinx has always offered these elements in some manner, but now we are making a concerted effort to bring our customers very defined and refined flows closely tied to our silicon, in what we call Targeted Design Platforms.
You can view our Targeted Design Platform approach as a pyramid (see Figure 1). Our Base Platform serves as the foundation layer of the pyramid. It is composed of our Virtex-6 and Spartan-6 FPGA silicon, our ISE Design Suite, and base development boards. The next layer above that is our Domain Specific Platform, in which we offer embedded, DSP, and connectivity IP, domain tools and reference designs, and FMC daughter cards that plug into our base boards. On top of that layer, we offer customers various Market Specific Platforms that are composed of communication, video, or AVB market specific IP, custom tools, and custom boards. If customers choose to use all these platforms for their designs, they can significantly reduce their overall design time and concentrate the majority of their design efforts on the value-added portions of their designs. Customers can, if they want, design every function in their designs from scratch, but most customers will certainly see the benefits of not reinventing the wheel and choose to concentrate on the value-added portions of their designs.
Smorgasbord approach changes
As part of the FPGA Targeted Design Platform approach we are also refining our tool flows to suit specific design disciplines. Traditionally we offered all our users one smorgasbord of tools and left it up to each user to figure out which tools match the tasks. However, we will soon be offering new domain-specific editions of our ISE Design Suite to help users pair tools with specific jobs while obtaining the number of tool licenses suitable to their budgets.
One of these editions, DSP Edition, bundles an improved System Generator for DSP with AccelDSP synthesis, DSP-specific IP running on top of the ISE design environment. The DSP Edition is primarily targeted at algorithm developers who are not HDL designers, but logic designers who do some amount of algorithm development will also find it useful. In addition, the Xilinx Software Development Kit (SDK) is a stand-alone tool, so if DSP Edition users want to do a bit of application software development to go with their algorithms they can add an SDK to the DSP Edition.
The move to a Targeted Design Platform approach makes sense when you consider how the FPGA and ASIC businesses have fared over the last 10 years.
Just 10 years ago, roughly 80 percent of Xilinx’s business came from the wired and wireless communications industry. Thus, when the dot-com bubble burst circa 2001, the communications business declined and in turn adversely affected our business along with the rest of the semiconductor industry. To ensure this wouldn’t happen again to Xilinx to the degree it happened then, in 2002 Xilinx quickly created vertical application groups in aerospace and defense, automotive, and Industrial, Scientific, and Medical (ISM), and in broadcast as well as in wired and wireless communications. Since then, these groups have helped Xilinx establish a much broader customer base (see Figure 2).
That investment in vertical groups has allowed us to diversify to the point where today wired and wireless communications only accounts for 44 percent of our business, while the balance is spread between ISM, A&D, data processing, commercial, and broadcasting. Indeed, today, despite the much broader scope of the economic downturn brought on by the mortgage crisis in the U.S., Xilinx has fared relatively well. In fact, we believe that this may be an opportunity for Xilinx and the rest of the FPGA space to more rapidly take share from the ASIC and ASSP business.
Why? While FPGAs and many other semiconductor sectors bounced back and grew out of the dot-com recession, one segment that did not was ASICs.
According to a 2007 report from research firm Gartner Dataquest, during the year 2000, the height of the dot-com boom, there were roughly 7,750 ASIC starts. But by 2005, there were only 3,623 – the number of design starts had been cut in half. Since 2005 ASIC design starts have continued to decline steadily, and most if not all research firms predict that ASICs will continue to decline over the foreseeable future.
Thus, when you take into account the historical impact the last recession had on ASICs and then consider that new logic IC manufacturing processes have become extremely more complex and exponentially more expensive, ASICs and ASSPs will only be feasible for an ever smaller number of applications. If you then consider that most economists predict this recession will be longer and deeper than the last recession, one has to wonder if the ASIC starts will again drop by half or more during this recession? Certainly, it isn’t a matter of if they will drop; it is a matter of how much they will drop/
However, to weather the storm and quickly seize new growth-market opportunities as they arise, electronics companies will still need to develop new products with logic ICs that differentiate their systems from the competition. FPGAs today are positioned better than ever to become the prominent logic devices in these new systems, as they offer companies the greatest mix of capacity, performance, processing, reprogrammability, and fast turnaround. FPGAs allow companies to quickly create new products with differentiated hardware as well as software functionality, allowing them to enter several new markets and quickly ramp to volume on those markets that prove most lucrative. They even allow design groups to add new functionality to their systems after they’ve released those products.
Where five years ago, the FPGA value proposition was primarily embraced by hardware design groups, today an ever-growing number of embedded software design groups, processor algorithm developers, and system architects are using FPGAs to create innovative designs and quickly get them to market. Why? Many of those companies who could no longer afford to build ASICs moved to buying ASSPs instead. But ASSPs offer limited programmability. That is, you can only differentiate your design in the software you program into the device. Your competitors likely have access to the same ASSP and can easily match or out-software your product. With an FPGA you can create unique hardware functionality as well as software functionality for your system and really differentiate your products.
We believe that our Targeted Design Platform approach will increase our customers’ productivity, and thus adoption of FPGA technology, and more rapidly take share away from both ASICs and ASSPs.
Certainly Xilinx and the rest of the FPGA industry is not immune to the effects of the worldwide economic recession. Before the downturn seemed imminent, Gartner predicted FPGA design starts would continue their growth rate and reach 90,000 for the year 2008. While we’ve yet to see new guidance from Gartner in regard to FPGA design starts, we believe over the course of this recession more individuals –embedded software designers, algorithm developers, and system architects as well as hardware design groups – will realize the programmable imperative and embrace the FPGA value proposition.
Moshe Gavrielov is president and CEO, Xilinx Inc. Moshe brings nearly 30 years of executive management and engineering experience with semiconductor and software companies to his new position at the helm of Xilinx. Most recently, he served as executive vice president and general manager of the fast-growing verification division at Cadence Design Systems. Before that, Moshe spent seven years as CEO of Verisity, Ltd., where he grew the company from a $4M start-up, taking it through its Initial Public Offering (IPO) in 2001, to a $70M publicly traded company, ultimately acquired by electronic design automation leader Cadence in 2005.
Previously, Moshe spent nearly ten years at LSI Logic Corp., where he served in a variety of executive management positions, including executive vice president for the $1.3B products group, senior vice president of international markets, general manager for Europe, and general manager of the ASIC division. Hw began his career in engineering and engineering management at National Semiconductor and Digital Equipment.
Moshe earned a bachelor's degree in electrical engineering and a master's degree in computer science from the Israel Institute of Technology (Technion) in Haifa, Israel. He possesses five patents.