Streamlining AI device design using AI and Innovator3D IC Integrator

Designing AI and Hyperescaler devices using advanced heterogeneous integration platforms, such as those from the Foundries and OSATs, and require the design team to invest considerable focus and effort on co-design optimization across the design hierarchy in order to realize their PPA & cost goals in a predictable resource efficient manner. This co-design optimization is commonly known as STCO, or System-Technology-Co-Optimization, a term and methodology created by the World leading research organization IMEC located in Leuven Belgium.

AI-powered STCO-related activities

With emerging AI and Hyperscaler devices containing multiple chiplets as well as high bandwidth memory stacks performing many of the STCO tasks necessary to achieve PPA and cost can be resource intensive and time-consuming and can often not achieve optimal results. Two such STCO-related activities are now AI-powered in Innovator3D IC Integrator as part of the 2604 release and deliver superior results in an efficient predictable manner.

1. The first AI-powered area is assigning and optimizing the power and ground delivery paths between the package substrate or interposer power delivery bump/balls and the chiplets power and ground microbumps. AI now drives power and ground substrate to chiplet assignments that minimize metal interconnect paths for optimal power delivery with minimal loss and metal route real estate usage while ensuring that the chiplet microbump/pin power requirement loads do not exceed the substrate/interposer bump/balls power delivery capacity.
2. The second AI-powered area is around the creation or optimized signal interface group paths that provide the design team with the capability to plan, define and optimize route channels in order to minimize congestion and metal layer requirements. A third AI driven capability is around the use of CoPilot technology to query the systems documentation using natural language queries. This can save not only precious time but also designer frustration when looking for guidance on performing a task or function.

For more information about what’s new in 2604 read this blog.