By Stelios Diamantidis, Sr. Director and Distinguished Architect, Synopsys Autonomous Design Solutions
The importance of semiconductors to the electronics industry has never been greater than it has been in recent years. Amidst the growth in technologies such as artificial intelligence (AI), 5G, the internet of things (IoT), and every company wanting to embrace this new world of innovation, global chip supply demands have placed significant pressure on the semiconductor industry. But as Synopsys CEO and Chairman Aart de Geus has made clear in recent media interviews, technology points the way forward.
While the industry has been hard at work on scoping different solutions, the fundamental problem lies in the imbalance in manufacturing capacity. With the pronounced lack of process nodes built on legacy silicon technologies, the market is awash with nodes in the mid-range of between 12 and 16 nanometers (nm). As a result, as many as 25 million wafers (10% of global capacity) go unused every year.
But what if we were to move these older designs to different process nodes?
Synopsys is proving that silicon remastering, a new AI-driven design framework in which AI reshapes old chips into new products, has the potential to transform the global chip supply chain for the better, easing the woes that manufacturers in areas including automotive and consumer electronics have faced in recent months.
De Geus’ keynote on silicon remastering at the annual ISSCC conference titled, “Catalysts of the Impossible: Silicon, Software, and Smarts for the Era of SysMoore,” was covered by top journalists and analysts at Forbes, Cambrian AI Research, and The Next Platform. Read on to learn about his thoughts on silicon remastering, the need to redesign old chips, and why using AI for chip design will be critical for tomorrow’s capacity.
As more companies start to design their own chips in-house, silicon technology will be used by more industries. Those that already do so, like the automotive industry, will require more of it. While semiconductor products were designed specifically for legacy silicon technologies, retargeting older chips using AI offers a way to move chip designs between nodes and absorb the market’s excess capacity.
At Synopsys, our breakthrough started as a simple thought experiment: Could we take a design created using one silicon technology and migrate it to another while maintaining the same level of quality? A round of experimentation showed that we could. In fact, we moved a 14nm node to 10nm with excellent results.
An article in Forbes cited de Geus’ keynote address, stressing how adopting this AI-led approach would effectively bring 25 million new wafers into the picture.
“Within five years, we will see a transformation of the global chip supply chain that will better facilitate the use of capacity, and we believe silicon remastering will be critical technology.”
We use AI algorithms to help re-optimize existing chips for different nodes — in a matter of weeks as opposed to months. In an interview with Karl Freund at Cambrian AI Research, de Geus talked about the many factors that come into play:
“One thing that will be increasingly necessary is to design chips in a way that they can be remapped at some point to a newer technology without having to start everything from scratch … Technology can already design things better than the traditional process.”
It’s no secret that the overall semiconductor design process needs to be more adaptable for design teams to absorb excess capacity. This goes beyond re-optimizing an existing design to a new node and points to a rethink of the way that chips are designed. Upgrading to a new node increases capacity and optimizes performance. Additionally, the design at hand gains the advantage of learning directly from the original design and an increased lifespan. Ultimately, this will inform the goal of designing new chips that can be readily remastered.
Silicon remastering is a perfect example of AI technology delivering a clear business solution. Not only does it save an immense amount of time, but it also helps customers set aside hundreds of millions of dollars (if not billions) that would otherwise be spent on a less efficient process or lost due to the disruption of the chip shortage.
The larger industry has started questioning the need to spend a fortune on building new fabs for old technology, when one can spend a lot less by employing AI to do the heavy lifting of migrating old (but good) designs to a modern equivalent. Automation tools simulate and verify designs before they head to the fab, rendering the chip in digital format. Assuming the design becomes a reality, AI technology can assimilate the original and re-engineer it for a new fab process. When chips are designed with remastering in mind, AI will also be able to deliver spin-off designs as well as complete remakes.
The early results of silicon remastering are very encouraging. Companies such as Intel, as discussed in a Forbes article, are already engaging with us in conjunction with its IDM 2.0 chip fab and foundry service expansion.
“Synopsys is a key design partner in our IFS Ecosystem Alliance. We are excited about the potential of technologies like silicon remastering to help Intel Foundry Services customers optimize their designs for PPA (Power, Performance, Area).” – Bob Brennan, Vice President of Intel Foundry Services (IFS)
With the appetite for chips across industries set to grow, we will need to find a way to live with and work around periodic chip shortages. Remastering represents an immediate and long-term solution because it creates both physical capacity and a deeper understanding of the design. Moving a design from an old node to a new one frees up capacity on that old node. The process teaches us about the design, tools, and similarities between technologies.
“The confluence of Moore’s Law ambitions in transistor design and now packaging coupled to systemic complexity that together will bring about a 1,000X increase in compute across devices and systems of all kinds and lead to a “smart everything” world.”
The article went on to explain that the cumulative nature of these advances is not additive, but instead multiplicative. According to de Geus, productivity gains from chip design tools have been growing at a log scale:
“The hand that develops the computer on which EDA is written can help develop the next computer to write better EDA, and so on. That circle has brought about exponential achievements. So often we say that success is the sum of our efforts. No, it’s not. It is the product of our efforts. A single zero, and we all sink. Great collaboration, and we all soar.”
Going forward, a key challenge to truly realize silicon remastering will be adapting to the “just-in-time” mindset of manufacturers like those in the automotive space. Solving this will require a joint effort from the larger industry with big implications. Remastering has already gone a long way to rebalance the world’s skewed manufacturing capacity, and we expect this trajectory to continue in years to come.
Watch Aart’s full keynote address on silicon remastering:
Catch up on other related blog posts: