What can AI actually do in 3D IC design today — and what's still science fiction?

What you’ll learn…

(01:25) Sudarshan's role leading global 3D IC software R&D at Siemens

(02:30) The real starting challenges of 3D IC design — and why there's no playbook yet

(05:25) Breaking down organizational silos between silicon, packaging, and mechanical teams

(05:49) Why semiconductor AI must be domain-specific, not just a chatbot

(10:34) AI as an engineering co-pilot for design space exploration

(13:38) Beyond design: AI's impact on materials, yield, and reliability prediction

(17:41) The shift toward AI-native, multiphysics co-optimization workflows

(21:16) Ethical pitfalls: explainability, data governance, and IP protection

(24:50) AI's role in the global, geopolitical semiconductor landscape

(27:48) How Siemens is building this with HEEDS and the Fuse EDA AI system

More about this episode…

In this episode of the Siemens 3D IC Podcast, host Tova Levy speaks with Sudarshan Deo, Software Engineering Manager for 3D IC at Siemens, about what AI can — and can't — actually do in 3D IC design today, and why the hype rarely matches the reality on the ground.

Sudarshan breaks down the biggest misconception in the field: that AI is either a magic box that will design chips end-to-end, or just hype dressed up as a buzzword. The truth, he argues, sits in between — semiconductor AI has to be domain-specific, physics-aware, and explainable, not a general-purpose chatbot. He walks through where AI is already delivering value, from narrowing design space exploration in chiplet partitioning and floor planning, to identifying promising materials and process combinations, to flagging reliability risks like thermal cycling and stress before they become expensive late-stage fixes.

The conversation also covers the organizational side — how 3D IC forces silicon, packaging, and mechanical teams out of their silos — and the ethical considerations that come with deeper AI adoption: explainability, data governance, and IP protection across foundries. Sudarshan closes by outlining Siemens' own approach, using Simcenter HEEDS for design space exploration and the Fuse EDA AI system to orchestrate engineering workflows end to end, while making clear that none of this replaces human judgment — it just changes how fast and how far that judgment can reach.

Ideal for: 3D IC and chiplet design engineers, EDA tool users, AI/ML practitioners in semiconductor workflows, packaging and reliability engineers, and anyone evaluating where AI genuinely helps versus where it's overhyped in chip design.

Connect with Tova Levy

Website

Connect with Sudarshan Deo

Siemens EDA