In the wake of the Spectre and Meltdown, it seems odd to write a post making the point I hope to make today. However as hard as it may be to believe, we truly are entering the golden era of the semiconductor industry.
I recall many personal conversations with Broadcom co-found Henry Samueli, where he would re-enforce his noted statement that there would be no more silicon/semiconductor startups. His conviction was because of how capital intensive it is to be a semiconductor company. In reality, we are seeing a semiconductor industry renaissance which is driven by a continuing increase of semiconductor performance in categories like cloud, automotive, AI, machine learning, etc.
This renaissance is partly driven by a number of new semiconductor/silicon startups coming from many parts of the world, not just the US, but also the combination of ARM making their IP more diverse and more turnkey. There are also more foundries looking for new business to fill their factories and offering a range of competitive offerings for upstarts. As interesting as watching a number of semiconductor startups bloom, what is most interesting to me in this golden era is the approach each company is taking to unique designs of their architectures.
In the past, there was a great deal of similarity in semiconductor architectures. This was during the speed race. Every company making processors was in a race to make the fastest. During that race, the focus was on the transistor design and the aggressive pursuit of Moore’s Law. But as I mentioned, the designs of the architectures themselves were quite similar. Interestingly, this is part of the pathway that led us to Spectre and Meltdown. The commonality in all CPU designs which led to Spectre and Meltdown was Speculative Emulation. Which in its own right is a very clever approach to speed up a processor, and an equally clever hack to exploit. But this simple borrowing of an architecture design to focus more on performance may very well be on its way out.
While the first big semiconductor era focused on speed, this next era will focus on efficiency. Nearly everyone designing CPU/GPU/memory/ASICs/Co-processors/ISPs, and most other types of silicon is now focusing on efficiency over performance. This is where, I believe, we will see some fascinating and creative engineering in the area of architectural design. Specifically, where we used to see many similarities in the architectural designs of Intel, AMD, Samsung, Qualcomm, Apple, etc., we will now start to see more originality from all the big names making semiconductors.
One area where we may see the most divergence is from companies like Apple, Samsung, and others who design chips mostly for use in their own products. A trend that we may see more companies follow like Microsoft, Google, Amazon, etc., over time. These companies have the most to gain from increased differentiation of their products and that is where architectural originality will ultimately benefit a company.
The best example I can think of today to highlight this trend is Apple. I have written extensively about Apple’s silicon design prowess, but the main point about unique architectures is emphasized with Apple’s strategy if you look at it this way: Apple designs processors that are totally useless to any other company but infinitely valuable to their product experiences. Apple uniquely designs their chipsets in a way that can only be used in their products because it is purposefully designed for their hardware and software exclusively.
From a competitive standpoint, Apple’s silicon strategy of highly customizing silicon to their hardware is one of the reasons the experience with iPhones, iPads, AirPods, and Apple Watch, is highly differentiated from the competition. Therefore, it stands to reason, that any company truly looking to set itself apart from the pack will, as often as possible, look to do more customization of their silicon solutions vs. use the same off the shelf parts that any competitor can use as well.
This point can extend to cloud providers as we will likely see Amazon, Google, and Microsoft start to make some custom silicon for their server hardware. We may similarly see more automotive companies add highly customized chipsets in their automobiles, and in a range of consumer devices as we move to personalized AI platforms that may exist in products outside of our smartphones like computers in our ears or on our faces.
As all companies look to increase their differentiation, I’m convinced that will increasingly lead to highly tuned and customized semiconductors. While not all companies may design them on their own, you may see Qualcomm, and Intel, do more of what AMD has done with their custom semiconductor division that helps companies customize solutions based on AMD architectures. Both the XBOX and PlayStation use custom semi-solutions designed in conjunction with AMD.
In a few decades, I believe, we will look back not at the 80-90’s as the semiconductor golden era but the 2020-2030s where we saw true innovation in design and architectures that push the boundaries of the science, and may even lead to a few new critical breakthroughs.