Skip to content
0MESubscribe
Chips18 Nov 20262 min read

Chipmakers Rethink Hardware Design to Tackle Heavy AI Workloads

Semiconductor leaders highlight a shift toward co-design, 3D chiplets, and secure data links as AI workloads outgrow traditional hardware methods.

Listen to this storyAbout 3 min, read by your device
Picture made with AI

Semiconductor engineering experts are rethinking how chips are built and deployed as artificial intelligence models place unprecedented demands on hardware. Traditional methods of boosting performance, such as simple voltage scaling, are no longer enough to meet energy and thermal limits.

Rethinking Chip Architecture and Memory

Engineering teams from across the industry are addressing the physical bottlenecks of computing. Synopsys experts point out that handling demanding AI workloads requires extreme co-design. This means engineers must design silicon, memory, interface IP, cooling, and power delivery together as a single unified system rather than as separate parts.

At the same time, the industry is moving away from single monolithic chips toward heterogeneous setups. Siemens specialists note that teams must navigate the complexities of AI-driven 3D integrated circuits and chiplets. Meanwhile, engineers at Rambus and Cadence focus on memory challenges. They explain that longer AI contexts, multimodalMultimodalAble to work with more than one kind of input, such as text, images and sound. Open in glossary models, and autonomous agentsAutonomous agentsAutonomous agents are software programs designed to make decisions and perform tasks independently. Open in glossary are shifting memory pressure, requiring hardware designers to tailor memory hierarchies to specific data streams, reuse habits, and latencyLatencyThe delay between a request being sent and a response being received. Open in glossary limits.

Securing Data and Moving to the Edge

As workloads spread across interconnected hardware, security becomes critical. Experts at Cadence highlight the growing need to build trust across physical links like PCIePCIeA standard hardware interface used to connect high-speed components inside a computer. Open in glossary, which now carry proprietary models, cryptographic keys, and confidential calculations.

Hardware strategies are also shifting at the edge. Arm specialists describe how lightweight smart glasses can run gaze-based tracking locally using fast on-device AI, while passing heavier workloads to larger computing systems. Arm engineers also point to growing hardware adoption of 8-bit floating-point data formats to keep models fast and efficient. Across the sector, leaders from SEMI and research hub imec stress that deeper global collaboration will be vital as chip designs grow more intricate.

How does this story land?

One tap per reader.

Quick poll

What is the biggest barrier facing next-generation AI chip designs?

One tap. No sign-up.

Ask this story

Still have a question?

Answers use only what is in this story. They are written by AI and can be wrong.

Sources

  1. Semiconductor Engineering, Blog Review: Nov. 18
Companies in this storyArmCadenceSynopsysWords explainedAutonomous agentsMultimodalLatencyPCIe
Up next in ChipsWhy AI agents in chip design may favor deep-pocketed giants2 min read

Keep reading

All Chips stories