Predicting The Future For Semiconductors

The number of new technologies and research directions means that there is plenty of uncertainty about the future of semiconductors.


Is it possible to predict the future? Of course not. We all make projections of what happened in the past, where they are now, and the implications for the future. We bias that in various ways and think we are making some astounding revelation, which is highly unlikely to become true. Of course, by luck, some people get it right and they are bestowed with grand accolades and awards. The likelihood of them ever getting anything else right is highly unlikely.

Or is it? How much of a role does luck play in this? It’s easier to assess if we turn it around. There are some people who seem to be very unlucky. I believe this is partly because they make bad decisions that increase the likelihood of bad things happening to them. I choose not to be impacted by hurricanes, which happen very frequently and somewhat predictably, and getting worse because of global warming, so I do not live in the Gulf states. I do accept the risk that a Cascadia subduction plate slippage could destroy the infrastructure where I live and cause untold damage and possibly loss of life. This happens once every 500 years or so, and I have taken reasonable steps to prepare for this event.

But what about semiconductors? In the past, many have talked about the end of Moore’s Law and been wrong. While it is more likely to be true this time around, it is not really relevant. What’s more important is a discussion about how a system is going to be better today than it was yesterday, because the industry is certainly facing a big storm. The lucky ones will be those who can navigate it, survive, and reap the benefits.

That requires companies to be nimble and adaptable to change. Will they continue to pack more transistors onto a single die, or will they deploy multiple dies connected together in a package? Will they explore new architectures or question decisions that were made decades ago? Will they explore new models of computation to deal with the evolving problem types that are transforming from being control-dominated to ones that are dataflow-oriented? Every one of these changes involves risk, but the one thing they all have in common is that they require an open mind and innovation.

In the past, change often has been driven by a fresh set of innovators, and they often attempt things that industry veterans will retort, ‘Tried that, didn’t work.’ But things have changed, and things that did not work in the past may be the new way forward. Many innovations were ahead of their time, and that often means the need was not large enough in the past and perhaps it is now.

I once claimed we could learn nothing of importance by looking at the past where semiconductors were concerned. I was wrong. Many things that were once abandoned are being revitalized. Many ideas that were too expensive at the time are now showing a lot more promise. A good example is memory. Alternative memory technologies have existed for decades, and they always were inferior in one manner or another. In many cases, the disadvantages were created because of other decisions that were made. For example, optimizing process technology for logic means it is not ideal for memory or analog or a number of other functions. But pressures are building. Flash and DRAM are having increasing problems scaling. A small change could make alternative memories a lot more attractive, if only we would give them a chance. Then, when they see an equal amount of investment compared to the incumbents, they may provide the breakthrough necessary for all manner of other innovations.

The impact of one decision can influence others. What if we gave up having notions of a single contiguous memory space as required by a von Neumann architecture? Sure, software would take a huge hit, but it could lead to much better, more efficient software that is amenable to a lot more automation than it is today. With a more analyzable software base, it would become easier for software engineers to create hardware, and that would open a whole new world of creativity. Most people are naysayers when it comes to software-driven high-level synthesis, because it is unlikely to create hardware as good as, or better than hand crafted. But what if there were 10X as many hardware designers, each with 10X more productivity?

The semiconductor industry has been chugging along without any fundamental innovation for a long time. There has been incremental innovation everywhere and that is what has driven most of the progress. In AI, it took a change from rule-based systems to statistical algorithms to enable the dramatic growth that we see today. Today, whole ecosystems have been created around that one innovation, and it has turned into a multi-trillion-dollar change.

I predict that companies will fail if they assume the future will look similar to the past four decades. What they need to do is innovate, and there are many ways to do that. Perhaps the best way is through the university system. It is too expensive for most companies to innovate because their expectation for results forces them into corners. Instead, companies should increase funding to universities, where many ideas can be explored quickly and cheaply. Maybe only 1 in 10 will be interesting, and 1 in 100 adoptable. Only 1 in 1 million will be transformational. Don’t expect them to create answers. Expect them to create the germs of innovation that can be incorporated into your future.

Luck happens when we create opportunities for success and when we nurture ideas, and those ideas shape the future.


Kumar Venkatramani says:

Thought provoking article, Brian! Thanks ;

One idea they may be worth exploring further or at least articulating is to list the number of axis on which current innovation is happening and the scale at which these technologies are expecting to change.
(Scaling transistors, multi-die, glass substrates, optical interconnect, EUV) or even software driven design, AI based design, die/package/substrate codesign and try and quantify the state of the art and proposed innovation today

Leave a Reply

(Note: This name will be displayed publicly)