Industry Insights: How Collaboration Will Accelerate Adoption of Multi-Die Systems

Expanding the Use of Multi-Die Systems

Multi-die systems have enabled chip designers to reduce both risk and time to market. For this reason, there’s been a notable expansion in the past two to three years, which is a far cry from the sporadic use across the industry in the past. During the panel, Samsung’s Park weighed in on key segments emerging in the multi-die system landscape.

“Adoption started mainly from high-performance computing (HPC),” said Park. “Now, we have folks in the mobile and automotive units that are seriously thinking of bringing those multi-die systems into those application spaces. What it really means is that the portfolio of technology and adoption of technology is expanding very fast. As a manufacturer and enabler in the market, we are really happy to see the ecosystem of multi-die systems expanding.”

Ansys’ Becer noted substantial growth in 3DIC design over the last year, forecasting even more growth in the future. He also touched on heterogeneous integration, a significant driver for adopting multi-die systems.

“The forecast for 3DIC is continuing to accelerate," said Becer. "In fact, we are seeing around a 3X increase in the number of 3DIC design starts this year versus last year, which means 3DIC is here to stay. One of the major benefits that influences further adoption is enabling heterogeneous integration. You can choose the best technology for each chiplet, meaning that you can bring in an analog, mixed-signal, or RF chiplet or you can use an older, more mature, or lower power node.”

Synopsys’ Kapoor echoed the sentiment of rapid growth within multi-die systems use, noting that Synopsys has seen a definitive increase in the usage of multi-die system designs, currently tracking over 100 multi-die system projects in the industry, spanning various package types and applications.

Exploring the Most Promising Applications

With compute demands growing faster than ever, the applications of multi-die systems are spreading to a few main key segments. During the panel, Bosch’s Schaffert touched on the automotive segment as one of the areas with the most promise for multi-die adoption.

“Looking at the vehicles of today and tomorrow, we will see a lot more software inside the vehicle," said Schaffert. "Our dilemma in the industry is that we have a mid-volume segment in the automotive industry, which means we cannot afford to develop a unique chip for each and every vehicle. Therefore, we are really looking forward to adapting this chip technology into the vehicles to have the reuse from other industries to get access to the very latest and greatest technology.”

In addition to the automotive segment, Synopsys’ Kapoor shared insights about applications in AI. Large language models (LLMs) require higher compute power to move data across the data center, a considerable driver for multi-die adoption.

“We talk about all these new AI applications and large language models like ChatGPT, which have large demands," Kapoor noted. "The workloads that they handle are huge. There’s also an increasing cost in maintenance and energy efficiency to move all this information across the data center. There’s a need to refactor this compute problem by looking at domain-specific architectures coupled with high-density memories like HPMS and bringing them all together in a small package. In other words, a multi-die system.”

Multi-die systems also show promise in addressing a pain point within the mobile market: high costs. During the panel, Ansys’ Becer explained why mobile markets stand to benefit from the adoption of multi-die systems.

“The mobile market is driven by the smartphone factor and low power requirements," he said. "We are very rapidly seeing a mobile market trying to implement a multi-die system because of cost reasons. They want to break out different functions into the advanced node and the more economical node. And they mix and match these pieces. So, it even provides a cost benefit to the mobile market."

Addressing Challenges in Multi-Die Systems

Challenges with integrating multi-die systems range from packaging optimization to high temperatures. Intel’s Immaneni and Ansys’ Becer provided insight into both issues during the panel. 

“The burden of these multi-die systems lies much more in the co-design and architecture than the execution and building," said Immaneni. "You’re bringing these IPs that are built on various process nodes, so you need to pick the right packaging technology. You also need to pick the right board technology." 

Beyond the industry’s need to come together and create standards for optimizing multi-die systems, other factors such as thermal challenges need to be addressed. Ansys’ Becer gave some insight into the risks of thermal stress.

He explained: "Multiple dies in the system create thermal challenges, and high temperatures are creating structural challenges, with possible disconnects in bumps due to thermal stress. High-speed digital signals for die-to-die communication through interposer require much higher fidelity electromagnetic aware extraction for more accurate signal integrity analysis. So, there are a lot of multi-physics challenges that are in play."