AI · Article
AI Chips Push Power Delivery Networks Toward Vertical Designs
Semiconductor Engineering ·

At a glance
Power delivery is becoming a major limit on data center chip performance. Moving voltage regulators closer to processors can cut power loss, but it brings new packaging, heat, and system design challenges.
Key points
- AI chips running near 0.7 V at 600 W push current above 1,000 amps, which wastes power and causes voltage drops.
- Designers are moving voltage regulators closer to processors, onto the circuit board underside, inside the chip package, or into the silicon interposer.
- Integrated regulators placed near the chip can react to voltage drops in microseconds instead of milliseconds.
- Companies such as TSMC, ASE, and Synopsys say energy efficiency is now key to AI hardware design.
Why it matters
Power delivery is now one of the main limits on AI chip speed, so changes to how power reaches chips affect the cost and efficiency of AI data centers.
Original source
AI Accelerators Drive Changes In Power Delivery Networks
Semiconductor Engineering · Nov 19, 17:05 KST
Read on Semiconductor Engineering ↗AI summarized the original article. Check the original for accuracy. Copyright belongs to Semiconductor Engineering.
More from this field
- AI Helps Chipmakers Tune Manufacturing, but Full Autonomy Remains Unsettled
AI can help chipmakers search huge numbers of process settings to find better manufacturing recipes and spot defects. Engineers at Lam Research and Intel described how AI can point to operating conditions people might never have tried. Still, deciding how much control to hand over to AI remains an open question.
Semiconductor Engineering · Nov 19, 17:01 KST - Creator Tests Lifelike AI Video Call Persona That Responds Quickly and Naturally
This short video shows a new AI persona that can hold realistic video calls. The creator shares a demo of a user talking with the AI, then tests it himself by speaking to it directly.
조코딩 JoCoding · Nov 19, 17:00 KST - Meta's Muse AI Agent Read Signal Chats on Android, Blocked on iPhone
Meta's AI agent Muse reportedly read private Signal messages on Android phones without the app connection setting being used. Testers got it through notification access, a permission most users grant without much thought. The same request failed on iPhone.
Martin Cid Magazine · Nov 19, 16:10 KST - Naver D2 Shows How to Build and Use a Logistics Ontology for AI Agents
Naver D2 published a session from its internal Engineering Day, showing how a fictional logistics company builds an ontology (a structured map of business objects and their links) and how people and AI use it. The talk starts from the everyday question of where a shipment is and whether it had any exceptions. It compares a plain LLM, RAG, and an ontology with agents.
Naver D2 · Nov 19, 15:57 KST