Macro trends defining the future of semiconductors

3 min read

Jim Costello, the former CEO of Cadence, talks about the future of semiconductors.

In an interview with Anari AI, Joe Costello, the ex-CEO of Cadence, who recently joined the start-up, which specialises in the development of personalised AI Systems, as a Board Advisor talks at about what he sees as some of the key future trends in the semiconductor industry.

The interview started with Costello looking at what he thought were the most important macro trends that are defining the future of the semiconductor industry.

According to Costello, “The semiconductor industry will continue to be the foundation of our most significant technology advances over the next few decades, but the future will be defined by different problems and constraints.”

The key macro trends that Costello sees in the next decade include the move from general-purpose to domain-specific compute architectures.

“Semiconductor companies today often have as many or more software engineers working at their companies, but despite this huge and ever-increasing investment, the software is still seen as a necessary evil/expense that must be done to get your chips sold.

“Traditionally semiconductor companies spend the least amount possible for the software and software support since they see the software as a freebie that is given away.

“The next decade, especially given the new domain-specific architectures, will require the boundaries between software and hardware to be eliminated and the full system solution to take the forefront for semiconductors. Systems are the perfect balance of software and hardware, and semiconductor products are systems.”

Costello then went on to identify fibre/photonics which he described as the foundation for all future high-speed communication. “It’ll be inevitable that we will see the world of photonics and electronics come together for the next generation of high-performance chips.”

Turning to supply chain issues Costello expressed concerns as to how the industry would look to solve the semiconductor supply chain issues in the coming decade especially in light of what has become an increasingly fragmented and hostile geopolitical environment? As he pointed out the semiconductor industry has thrived in a globally free economy, so how it continues to make progress would be a significant challenge.

Next Costello spoke about the transition from general-purpose to domain-specific architectures and why general-purpose solutions weren’t working anymore.

“The Turing Lecture delivered by Professors John Hennessy and David Patterson in June of 2018 marked a clear turning point in the semiconductor industry,” said Costello. “The previous 40 years was the era of Moore’s Law where successive generations of chips could pack more transistors on a single chip and thereby deliver ever-increasing processing power for almost all applications. But Moore’s law (and Dennard scaling for power) no longer holds and there must be a new approach to the design of chips for high-performance computing applications. Domain-specific architectures will lead the way for the semiconductor industry to continue to deliver ever-increasing power to solve the world’s biggest problems.”

When it comes to the Cloud and its importance in creating next-generation hardware, he said that it had, “Changed the game and democratised computing. Before the cloud, only large rich companies who could afford to build enormous data centres could attack problems requiring huge compute throughput. With the cloud, a single engineer can have access to more compute power than engineers at the largest, richest, and most powerful companies in the world. We still haven’t fully unleashed the potential of this shift. We humans haven’t yet fully digested this and exploited its full potential.”

As to the impact of artificial intelligence on semiconductors Costello suggested that AI/ML was perhaps the most important IT development of this decade (or more).

“The ability to make our machines learning machines that can outperform the best humans at most tasks is going to transform the world as we know it. When perhaps the greatest Go player of all time lost to Google’s DeepMind in 2017, I cried because it marked a watershed moment for human beings that I don’t think we yet comprehend. It isn’t a bad thing but it is an enormous shift that we must acknowledge and embrace. As with any other profound wave in IT, the semiconductor industry will lay the foundation for delivering on the full promise of AI/ML with generations of advancements in domain-specific architectures and the tools to build them.”

Asked about the broader ecosystem Costello was then asked why it was so important to follow in the future?

“All of the greatest things that have been accomplished by human beings through the ages have been accomplished by combining the energy of large numbers of humans and human minds.

“Creating an ecosystem where large numbers of people can work together coherently will create far more progress than even vastly greater numbers working independently with no cohesive vision or mission. It is analogous to the difference between incandescent light and lasers. A 100-watt light bulb produces some light, but mostly heat and the incoherent light has some, but not great effect. A 100-watt laser where the light is coherent is immensely more powerful and efficient. Ecosystems are coherent human minds – they will always win.”

  • This is an edited version of an interview Joe Costello had with Anari AI