About the Group

A forum for the hard problems sitting between silicon and software.

The AI Software Interest Group brings together engineering and research leaders from across the semiconductor ecosystem to examine, in candid and pre-competitive settings, how artificial intelligence is reshaping the way software is designed, built, and operated. Each member contributes a perspective from a different corner of the industry — memory, design automation, systems, mobility, and beyond.

Our first round of internal presentations is a mapping exercise: surfacing the themes that matter most to the group before we commit them to the record. The summaries below describe the topics under discussion ahead of June 1st, shared without attribution while individual findings remain in review.

Leadership

The ASIG launched in Spring 2026 under the guidance of the following founding members
Chair Mandali Khalesi Vice PresidentAI Platform & Technology Renesas
Co-Chair Dr. Nicky Lu Chair/Founder Etron Technology

Together with GSA, they will help guide ASIG’s strategic direction and ensure its impact across the global ecosystem.


Topics Under Discussion

Sharing the topics under discussion as we evolve from member-centric case studies to crafting a shared vision for our industry.
01

Next-Generation Memory Architectures

Modern AI accelerators sit idle for 60–80% of wall-clock time waiting for data, making memory one of the binding constraints on deployable intelligence. ASIG is examining how HBMx, CXL 3.x memory pooling, and processing-in-memory architectures is shifting the focus from compute to memory-system co-design across cloud and edge.

Hardware
02

Vision-Centric Agentic Workflow for Factories

AI agents are moving beyond copilots to become active workflow participants on the factory floor. Combining process automation with cognitive AI can help handle complex unstructured data such as machine logs, shift notes, and domain knowledge. ASIG members are contributing case studies from pilot to production, focused on operational reliability required for end-to-end robotic data processing.

Manufacturing
03

Cybersecurity & Data Privacy

Industrial buyers are screening AI suppliers for clear trust posture: confidentiality, integrity, and compliance becoming core requirements as AI handles sensitive data and autonomous decisions. ASIG is framing privacy-preserving AI, and defenses against model theft and data poisoning across the full AI lifecycle, focused on industrial environments where breaches create business and operational risk.

Trust
04

Agentic Workflow for Semiconductor Design

Agentic AI systems that can plan, call tools, and execute inside EDA workflows are delivering 20–50% productivity gains in key areas such as place-and-route and verification. ASIG is exploring how the competitive question for design tools is shifting from “does it have AI?” to “is it a workspace an agent can usefully operate inside?”. APIs, observability, and supervision need a more rigorous discipline as their usage grows.

EDA
05

Hybrid Quantum / Classical AI Computing

Hybrid workflows that decompose problems into subjobs routed to the best-suited processor (CPU, GPU, or QPU) are now operational in national programs including Japan’s JHPC-Quantum, the US OLCF at Oak Ridge, and EuroHPC centers. ASIG is examining how semiconductor firms can build early organizational fluency through targeted pilots in combinatorial optimization, materials simulation, and multi-robot coordination.

Compute
06

Multi-Robot Orchestration & Planning

Coordinating cells, lines, and entire facilities of heterogeneous robots under a shared goal is emerging as the most strategically rich intersection of classical optimization, agentic AI, and hybrid quantum-classical methods. ASIG members are defining a vision for multi-robot orchestration where next generation models enable generalization across heterogenous robots, collapsing deployment and fault recovery timelines from weeks to hours.

Robotics
07

Frontline of Cyber-Physical Systems

Deploying AI in cyber-physical systems, where software intelligence coordinates tightly with sensing, computation, and actuation, demands deterministic behavior under real-world variability. Generative AI’s tolerance for plausible-but-wrong outputs is not viable in mission critical industrial workflows. ASIG is sharing practical lessons in safety engineering to define a best-practice framework for reliable AI in where latency, resilience, and regulatory compliance are non-negotiable.

Systems


Participating Companies

Members of the AI Software Interest Group
Etron Technology
Decloak
FPT
Cadence
Panasonic
Softbank
Renesas