Why the semiconductor industry needs less rigid walls and a new roadmap.
If the still evolving shift left has taught the semiconductor industry anything, it’s that nothing can be counted on to stay the same for very long. The basic proposition of doing more earlier in the design cycle to speed up time to market—and thereby shifting everything left in the linear flow—means that every silo that has been constructed over the past several decades needs to be rethought.
This kind of thinking pops up every decade or so in the design world. Silos are the most efficient ways of getting things done until something changes, and then they become more of a hindrance than a help. This is true in every industry, but it’s particularly true in the semiconductor world. The chip industry has proven over the past 50 years that even when you thought it had hit the limit, it could still squeeze another fraction of a penny out of the supply chain and wring one more process node out of existing technology.
What’s different now is that change is happening more quickly than ever before, and it’s happening in more places and in more markets. The whole IoT/IoE technology/business proposition is based on the ability to customize everything quickly, to build platforms and revamp them as needed, and to put them together in unique ways that can be partially pre-verified but with the capabilities of being partially customized or fully customized.
This kind of thinking would reduce established flows to chaos in normal times, but these are not normal times. The difficulty of continuing to shrink features is rising. EUV remains questionable—the power supply is still insufficient and uptime is inconsistent—while yields at advanced nodes remain low. Even Intel has been struggling with 14nm yields. Everyone agrees it gets much harder from here.
All of this has led to widespread experimentation with new types of packaging, the addition of new memory choices and more materials options. But what does this mean for a design-through-manufacturing flow when designs need to be customized, more steps are run concurrently, and markets that were never considered in the same sentence are suddenly connected. And now, rather than just confined to the chip, pieces of the PCB are being incorporated into the package, all of which has to be optimized for power, performance, physical effects, tested and verified, maybe even at the same time.
The success of every company in the semiconductor ecosystem requires a rethinking not just of the technology, but of the business of how to build the next generation of chips. In short, we need a new road map, and it’s one that cannot just be based on line widths and memory sizes. Almost every conference in the industry is focused on narrow slices of technology and how to optimize power or performance or the integration of particular components. The glaring hole is what is the best infrastructure for building these devices.
In the past, this was largely left to the chipmakers to decide. That doesn’t seem to be working anymore. Acquisitions of startups and a replenishment of new ideas with new startups is a sign of a healthy ecosystem. Acquisitions of large companies by other large companies is a sign that something fundamental needs to change and that the old way of doing things no longer is working.
What will a successful company look like in five years? How will it be structured? And how long will that structure last before it needs to be rethought and reconfigured? These are all questions that few people are asking today, but given the breadth and nuances of the chip industry it’s also something that companies are unlikely to figure out on their own. It’s time the best minds in the business world put their heads together to begin tackling these issues so the industry can get back on track and continue doing what it does best—figuring out new and more efficient ways to build the engines for the next few decades of technology growth.