中文 English

The Race To Design Larger Systems

The semiconductor industry finally is getting the chance to prove itself on a much bigger scale.

popularity

For more than a decade, tools vendors and design houses have been talking about leveraging their tools and expertise to help design systems of systems. They’re finally getting their chance.

The basic idea behind this strategy has always been that issues inside any electronic system—performance, power, signal integrity, area—have all been dealt all the way down to the sub-atomic level in chips. EDA companies can identify hot spots on chips long before the chips are manufactured. They can spot irregularities in wiring, and they can determine whether a design will yield sufficiently in manufacturing. They even can make predictions based on various use models about how long chips will hold up under harsh conditions, such as a logic chip in an autonomous vehicle in the desert sun. And design houses know how the various pieces interact on a level that carmakers aren’t even contemplating.

Until very recently, all of this tooling and expertise has met with only limited success outside of the chip market. The problem was largely a matter of siloes, and very few tools developed for chips (exceptions include computational fluid dynamics and wiring harness modeling) ever made a dent beyond the semiconductor world.

Three things have changed over the past few years to alter that equation significantly. One involves a fundamental business shift, with system companies such as Apple, Google, Amazon, Alibaba, Facebook and others now taking over the design of their own chips. Rather than focusing on cutting every last fraction of a penny out of the cost of chips, systems companies have started developing chips that make their devices or systems more competitive. That resulted in a wave of consolidation among fabless chipmakers, and that trend is likely to continue.

This has proven to be good news for makers of processor cores, whether those are from Arm, Synopsys, Cadence, some variant of RISC-V, or even embedded FPGAs, as well as memory makers and RTOSes from companies like Mentor. These are pretty much snap-in components that have been well characterized and silicon-proven.

But systems companies are pushing into a lot of new areas at once, and they have limited expertise in many of these areas. That presents a big opportunity for EDA and IP companies, as well as system houses, because they have the expertise to reach well beyond the silicon or the board to design, lay out, simulate, verify, package and test all of these components working together under a wide variety of use cases. In fact, they’ve been champing at the bit for this opportunity.

The second change involves the slowdown in device scaling, aka Moore’s Law, which means all of these companies now have to architect their way to better performance and lower power rather than just migrating to the next process node. So rather than paying hundreds of millions of dollars to design a chip that provides a 20% or lower improvement in performance, why not design a system that provides a 100X improvement? That’s exactly what’s going on, and it has fueled the biggest infusion of venture capital in semiconductor startups since the 1980s.

This development coincides somewhat serendipitously with the third change, which is an explosion in AI/ML and the push to more processing at the edge. This isn’t just a rehash of the client/server model from the 1980s. It’s more akin to a balancing act between data generated and pre-processed at the edge and data that is mined and acted upon in the cloud or any centralized compute system. In this mix, the data defines the architecture, and the architecture is designed around that to accelerate the data.

The tools and methodologies are similar, but the difference is this isn’t just about chip design. It’s about optimizing the flow of data rather than signals, and that can include everything from multi-directional read/write in memory and arrays of processors and memories all working together or separately.

Put all of these pieces together, and this creates one of the greatest opportunities for EDA, IP, system houses and even equipment makers to break out of the box. And this opportunity will only grow as more electronics are used to replace or enhance mechanical components.



Leave a Reply


(Note: This name will be displayed publicly)