Moore’s Law is Dead: So, Let’s Talk About the Future of SoCs

Sensors Insights by Jack Kang

By now, you’ve likely heard someone tell you: Moore’s Law is dead. To be sure, it is worthwhile to point out that the design aspect of Moore’s Law is – in most circumstances – very much alive. There are still techniques to be discovered to make the transistors smaller, for them to work faster, and to still put more of them in the same footprint. Rather, it is the economics behind Moore’s Law that has clearly reached its endpoint. Chips might continue to get smaller and faster. But if the economics mean that almost no one can afford to buy them, where exactly are we headed?

The upfront cost of a new design start has increased so exponentially that it has priced out the small to mid-sized company without tens of millions of dollars minimum in loose capital to justify an investment in custom silicon. This is leading to a logjam in many industries, i.e., those who need access to custom core IP to differentiate their product(s) simply cannot get it. They can’t justify the upfront cost, and neither can the traditional silicon providers.

The traditional IP providers are all chasing the next killer socket so that they can find enough volume to justify the development of a new chip. But what if the next killer socket doesn’t have a single design point that can be serviced with just a few chip designs? What happens when the market demands no longer fit the business model of the economic realities of Moore’s Law? Take, for example, the rapidly evolving internet of things (IoT) and artificial intelligence (AI) sectors.

 

Internet of Things & Artificial Intelligence

The internet of things (IoT) markets are comprised of billions of small devices, with this number only growing larger by the day. Gartner predicts there will be 13.5 billion connected “things” in the consumer sector by the end of 2020. Many of these devices are going to require intelligent computing. These “edge compute” devices naturally lead to a very fragmented market with large ambiguity in potential winners. No wonder the industry is having a tough time servicing the market.

Without the ability to differentiate their products via custom silicon to stand out in such a noisy industry, no system vendor has been able to claim victory. We believe that these system architects know what they physically need, but they also require a new approach to get custom chips at price points that make sense.

It’s no surprise artificial intelligence (AI) is booming. Yet despite all this talk, the growth of AI will come to a screeching halt if the semiconductor industry doesn’t get its act together. Let’s break it down to a simple fact: AI requires massive amounts of computing. The chips currently available can’t handle those processes. To churn out newer, more capable cores, silicon needs to be easily customized.

That’s a huge challenge because the market is cornered by a handful of companies. It has become nearly impossible for inventors to start new designs because of insanely expensive licensing, design, and implementation costs.

To get custom core IP into the hands of the companies and industries that need it most, there needs to be a fundamental shift in the way chips are manufactured and distributed. Reiterated, Moore’s Law is economically dead. Chips may still be getting smaller and faster but if nobody can afford them, the inability for companies to differentiate will continue.

While there are, perhaps, a few ways to solve this challenge, an open-source, SoC design might just be the change the semiconductor industry is looking for. It’s no secret that when it comes to getting ahold of custom core IP, a limiting factor for many organizations is price. And it’s also no surprise that, ultimately, this model will never be completely flipped on its head. The larger chip vendors will continue to support the traditional model of a robust production layout, with upfront costs in the tens of millions of dollars. 

 

RISC-V: A Viable Solution

One alternative gaining recognition/traction in development circles is the RISC-V instruction set. The free and open RISC-V ISA has gained support not only from academic partners but also from industry leaders such as Google, NVIDIA, Qualcomm, and Oracle. RISC-V allows for the customization that currently seems to be lacking, and the infrastructure necessary to address the otherwise stagnant designs that are inherently limiting. 

System architects know what they need – purpose-built, custom chips. Simple by nature, yet so challenging to achieve. So how do we get there? If the current semiconductor business model stays its course, the industry will continue to carry its top-heavy weight.

The Freedom U540 is described as the industry’s fastest RISC-V Processor and the only Linux-capable RISC-V SoC.
The Freedom U540 is described as the industry’s fastest RISC-V Processor and the only Linux-capable RISC-V SoC.
Freedom U540 SoC Block Diagram.
Freedom U540 SoC Block Diagram.

Thousands of system architects need alternative options for getting their hands on the tools they need to do their job well. RISC-V and SoCs such as the Freedom Unleashed 540 are just a couple of examples of where the industry can turn to, to give power and resources to the companies that need it most.  By taking an open-source approach and enhancing mass customization, Moore’s Law will concede to the future of open-source SoCs one chip at a time.

 

About the author

Jack Kang is the vice president of product and business development at SiFive, a provider of commercial RISC-V processor IP. Jack has a proven track record of driving cutting edge technologies and new products from inception to productization, previously having worked for companies such as NVIDIA and Marvell Semiconductor.