In the past two weeks, two separate experiences led me to reflect on Moore's Law and whether the tech industry needs to stay on the same pace of miniaturization in semiconductor process technology. Few applications can take advantage of the densest chips these days. And the latest processes have driven manufacturing costs through the roof. Indeed, progress, in this case, make end up hurting innovation.
So let's get back to the two factors that prompted this column. Two weeks back, I attended a Xilinx media event that both celebrated the company's 25th birthday and launched the new Virtex-6 and Spartan-6 family of FPGAs. The announcement marked the first time Xilinx had launched new families simultaneously in both product lines and moved both simultaneously to new processes (40,45-nm). The product families now are strikingly similar in architecture and both feature increased density and lower power. See Dylan McGrath's EE Times story for more product details.
It wasn't so much the product announcement but Xilinx President and CEO Moshe Gavrielov that prompted my thoughts. Of course Gavrielov provided the expected FPGA-centric message using a theme called "The Programmable Imperative." Check out this article and video for a first hand look. Gavrielov claims that a combination of design cost constraints, the need for connectivity, and the consumer's constantly changing demands are driving FPGAs with hardened high-speed I/O into a far broader set of applications than ever before.
Gavrielov painted a really bleak picture for the bulk of the semiconductor industry. And note that the bleakness had little to do with the current recession. Instead Gavrielov relied on the escalating cost of chip design and manufacturing to place Xilinx as one of few long term survivors in the IC space.
Gavrielov correctly points out that even top-tier companies such as Texas Instruments have moved to a fab-lite model and may be headed to a fabless future. It's Gavrielov's view of the second tier that really surprised me. He claims that the mid-tier companies that have successfully made products such as ASSPs for communication applications simply can't recoup their investment in new chip designs any longer. Moreover, Gavrielov claimed that venture money for semiconductor startups has disappeared -- presumably forever. Now remember that Gavrielov's mission is to sell FPGAs so you would expect a pitch that justified the category, but still these were strong predictions of a major change in the IC business.
I returned home from the Xilinx event and then read the column that Bill Schweber wrote for EE Times last week. Now Schweber, of Planet Analog fame, is an analog type but he's also a keen industry observer and never one to dismiss the digital innovations in our industry. Schweber's column does a great job of painting a bleak picture of the cost of continuing to ride Moore's Law. There is just no margin for error -- in terms of design or an economic glitches -- for companies that depend on prior-generation sales volumes to fund new developments. And right now even the fab guys are hurting with low utilization rates.
What I don't understand is why we can't put the speedster that is the industry on the Moore's Law down ramp in neutral or reverse. The fact is that those mid-tier companies could still effectively build their ASSPs on 90- or 130-nm processes. Sure those designs would feature a bit higher dynamic power consumption, although less static leakage loss. Fabs would get much longer to recover investments in new process lines. And designers would continue to have access to a broader range of ASSPs and even ASIC technology that design cost have pushed out of the mainstream range.
The fact is that outside of the handset and PC space, no ICs other than memory and FPGAs can take advantage of the latest processes. I'm sure Altera and Xilinx will both do quite well going forward. But pace of technology doesn't have to render the rest of the IC industry obsolete. The fact is that we need the innovation that has always come from mid-size and startup companies.