From Gordon to Geoffrey - Which Moore's the Law?

It was the best of times. Gordon Moore’s prophetic paper from 1965 did not predict the possibility of programmable logic, but in proclaiming that device densities were destined to grow at the startling rate of almost a factor of two every two years, it set the stage not only for the invention of FPGAs, but for their proliferation in a wide variety of applications. True to the law of Gordon Moore, programmable logic raced its way up the sophistication curve, fueled by the geometric increase in capability. During this time, the makers of FPGAs learned to listen to the voices of trusted advisors, their most sophisticated customers in the networking and telecommunications industry. Those voices said “ever bigger, ever faster,” which is the mantra of Moore’s law.

Traditionally, FPGA customers have valued high density, high pin counts, and high speed. They have typically placed a smaller premium on pricing, power, and ease of design. They have driven the FPGA vendors to produce flagship product lines with premium pricing such as Xilinx’s Virtex and Altera’s Stratix families.

It was the worst of times. The telecommunications and networking companies were suffering a serious drought, and their appetite for expensive FPGAs was at least temporarily sated. At the same time, it appeared that the bounty of Gordon Moore might be finally coming to an end after a forty year run. Exponentially increasing density had walked hand-in-hand with exponentially increasing development cost for each process generation, and the greater laws of finance and physics would inevitably show their might.

Geometries in FPGA have now reached the 90nm mark, and, for the first time, a new process node isn’t all good news. In the past, little compromise was required. Smaller geometries gave greater density, lower cost, higher speed, and lower power dissipation at lower supply voltages. Now, for the first time, designers are faced with a “choose your favorite two” situation. Leakage current causes power to actually increase at the 90nm node, and design compromises must be made to get around it. Many industry experts (and yes, we’ve all heard this several times before) are predicting an end to Moore’s law within the decade.

As Gordon’s law shows signs of running out of steam, another Moore is here to take his place in the FPGA environment. Geoffrey Moore’s books on adoption of new technologies by the market have become to many high-tech marketers what Gordon Moore’s prediction was to engineers: a treatise that puts their world into perspective over a relatively long period of time; a framework that puts their fast-changing environment into perspective and organizes it with some semblance of order. In “Crossing the Chasm” and “Inside the Tornado,” Moore describes the progression of new technologies penetrating the market, making their way from the bleeding-edge “innovators” to the progressive “early adopters,” across the “chasm” and onto “main street” where they become commonplace.

In his detailed analysis of the dynamics of the market at each stage and the buying behaviors of each of these groups, Moore describes almost exactly the situation seen in the FPGA market. As the technology spreads to a larger and more diverse customer base, the priorities and driving factors change. The nature of the product changes as well. The key success factors shift from performance and raw capability to packaging, ease-of-use, and completeness of the solution.

We asked Geoffrey Moore what he thought about the changing dynamics of the FPGA market. “At this point in the market,” says Moore, “you have to change who you’re listening to for market direction. Until now, the vendors have made their way by listening to the expert customers who are pushing the technology to the limits. As the technology now becomes more widely dispersed, however, they need to start listening more to the less sophisticated customers who value other aspects of the product such as ease of integration.” Moore says a common pitfall of suppliers is to continue to follow the leadership of their expert customers exclusively and miss out on the bigger opportunities as a result.

As applications such as consumer and automotive increase their adoption of FPGAs, we should see Geoffrey’s laws take the lead from Gordon’s. Business will be won or lost based on IP aimed at specific design problems and application areas, tools that are easy to use and integrate well with other parts of a specific product development process, and previously less important properties such as power, footprint, design security, reprogrammability, price, price, and price.

As the size of the audience increases, its sophistication also decreases. The induction of new customers into programmable logic means that there is a training barrier faced by newcomers to the game. “As their products reach Main Street,” says Moore, “marketers have to start listening less to the A students and more to the B and C students.” For FPGAs to be successfully adopted by design teams that don’t have years of experience in VHDL or Verilog design, the design process will need to be more approachable.

Already, we’re starting to see some FPGA-based projects driven by software engineers instead of hardware designers. With the advent of embedded soft-cores such as Xilinx’s MicroBlaze and Altera’s Nios, and with the drop-dead simple design tools supplied with them, even a fairly unsophisticated user can quickly and easily set up a working embedded processor complete with peripherals and RTOS running on an FPGA development board. From there, it is a simple step to have application software executing and the device performing its basic functions, all with virtually no involvement from hardware types. The first discontinuity in this process, in fact, comes when software modules need to be ported into VHDL or Verilog for acceleration in the “hardware” portion of the FPGA (implemented in the fabric as opposed to running in software on the embedded processor). If progress continues in C and C++ synthesis, this step, too, may eventually become nothing more than an alternative compilation method from a designer’s point of view.

So, if you see Moore’s law start to flounder, fear not. It’s a far better place we’re going, and a new Mr. Moore will be there to show us the way. While we probably won’t see FPGA development kits on the shelves at Wal-Mart in the near future, we should see a larger and more diverse customer base taking advantage of FPGAs than ever before. The capabilities and features that are developed to meet the demands of these emerging markets are likely to benefit us all.

Kevin Morris, FPGA and Programmable Logic Journal

May 4, 2004

[back to top]

Comments on this article? Send them to comments@fpgajournal.com