"EDA For AI, AI For EDA" — a New Era of System-Level Design

Xpeedic officially rolled out an all-new version of its EDA2026 product line at DAC 2026 today. Xpeedic EDA2026 delivers a system-wide upgrade across three signature capabilities — electro-thermal co-analysis, opto-electronic co-simulation, and EDA Agents — reinforcing Xpeedic's position as the System-Level EDA leader for AI hardware design.

Highlights at a Glance

• Electro-thermal co-analysis: Notus, Xpeedic's board-level simulation platform, now models two-way electro-thermal feedback; paired with mesh reuse and cluster-parallel computing, it cuts simulation time dramatically — validated on a real server board from xFusion.
• Opto-electronic co-simulation: ChannelExpert and Metis now work together to support ultra-high-frequency SI, cross-scale opto-electronic modeling, and thermal-stress coupling for CPO packaging — validated on a 2.5D packaging project with YIBU Semiconductor, boosting simulation scale 20x.
• EDA agents deployment: Built on the XAI platform, six EDA agents are now embedded across the STCO design flow, enabling multi-physics AI agent collaboration and chip-to-system end-to-end autonomous optimization — already deployed at scale in products like Lenovo's AI PCs.
• AI ecosystem partnerships: Announced alongside EDA2026 at DAC — the latest EDA Agent R&D results with Lenovo Group, and a strategic AI+EDA partnership with NovaStar for LED display-control chips.

I. Electro-Thermal Co-Analysis: Breaking Down the Wall Between Electrical and Thermal Simulation

As power draw on AI server boards climbs past 6 kW, current and heat in the power busbars increasingly feed back into each other — a coupled problem traditional tools can only solve one domain at a time, with engineers manually shuttling data between the two.

EDA2026 gives Notus two-way electro-thermal feedback modeling, so current and temperature fields solve together automatically instead of relying on manual handoffs. Combined with mesh reuse and cluster-parallel computing — which recalculates only the parts of a design that actually changed — large-scale board simulations that used to take days now take hours. For more complex 3D structures, the thermal-stress simulation platform Atlas, together with electro-thermal co-simulation in Boreas and Janus, extends this same approach to the package and chip level, so the whole electro-thermal chain, from chip to board, can be solved within a single toolset.

The approach has already been validated on a server board design from xFusion. In the company's words: " Notus's high-precision two-way electro-thermal feedback solved a busbar thermal-reliability challenge that's been a real headache for high-density board design. The efficiency gains have also given us solid support for reliability design in high-power delivery systems."

II. Opto-Electronic Co-Simulation: Cross-Scale, Full-Chain Simulation for the CPO Era

As co-packaged optics (CPO) moves toward volume production, simulating it means solving optical and electrical effects together at wildly different physical scales, while also accounting for how heat affects optical performance.

In EDA2026, ChannelExpert adds opto-electronic co-analysis, capturing both electrical and optical effects in a high-speed link at once. Metis now automatically bridges optical devices (micron-scale) and package structures (millimeter-scale) — no more manual stitching between models — while holding simulation speed steady even at higher frequencies.

The capability has been validated on an advanced packaging project with YIBU Semi: simulation scale improved 20x and computational efficiency improved 10x over conventional methods. YIBU Semiconductor's team put it this way: "Xpeedic's Metis let us run high-frequency signal, opto-electronic interface, and thermal-electrical coupling simulation together on a single platform, with results we could plug straight into link budgets and structural optimization — it cut a lot of time out of catching problems early in the design."

III. EDA Agents: From Rule-Driven to Data-Driven Design

EDA2026's core shift: Xpeedic's XAI platform now lets large AI models take over the repetitive, experience-driven work — modeling, parameter hunting, optimization — that engineers used to do by hand. The advanced packaging platform Metis has already deployed an AI agent that understands design intent and calls the right tools to run simulation on its own; the multi-physics simulation platform XEDS adds AI-driven parameter optimization that automatically finds parameter sets meeting design targets; and the board-level platform Notus can now use AI to predict thermal simulation results directly, without recalculating from scratch every time.

Xpeedic has pushed AI into physical-layer simulation — electrical and thermal — across the entire flow from chip to package to board to system, and that's exactly where the company has focused its investment in recent years and built a real edge.

The capability has already been validated across multiple customer projects, with simulation efficiency improved by as much as 70x — cutting work that used to take days down to minutes. In the collaboration with Lenovo Group, AI improved PCB design efficiency by more than 50% and simulation efficiency by more than 2x.

Pave the Way for System-Level Design in the AI Era

With EDA2026, Xpeedic has completed a systematic upgrade across three fronts — electro-thermal co-analysis, opto-electronic co-simulation, and AI multi-agent design — further cementing its position as the system design pathfinder for the AI era. Xpeedic says it will keep building around "EDA For AI, AI For EDA," deepening the fusion of its STCO technology roadmap with EDA agent capability, to give system-level, fully intelligent support to the design of high-speed, high-frequency electronics and AI hardware infrastructure.