Scaling AI with Wide, Parallel Interconnects

In this episode, Jun Kawaguchi, a marketing executive at Winbond, will explore the emerging role of Flash memory in client AI, where AI processing happens at the client location rather than relying primarily on centralized servers. The discussion will look at how this shift affects memory requirements and the role NOR Flash can play as more AI workloads move closer to the user, with a focus on what this means for developers and system designers.

AI, space and defence are creating opportunities for Sivers Semiconductors' high-performance photonics and wireless businesses. In this episode, Sivers CEO Vickram Vathulya discusses the company's technology, manufacturing and business strategy.

In this episode, Sally talks to Manoj Koul about how workload-first, programmable platforms are reshaping compute for Physical AI.

As AI clusters scale, network reliability is becoming critical. In this episode, Credo’s Don Barnetson explains to Sally why the company is pursuing a “wide and slow” optical architecture, how its 2D emitter arrays could deliver 100× scalability, and how Active Light Cables could bridge the gap between copper and optics.

In this podcast, we chat with to Majestic Labs’ Sha Rabii about why AI's shift toward inference, long context, and agentic workloads could require a fundamentally different compute-memory architecture.

In this podcast, Sally talks to Amir BarNiv, VP of Strategic Marketing for Automotive Ethernet at Infineon. As AI becomes an increasingly important part of the software defined vehicle, a different architecture is needed in the car. So what does the AI-powered vehicle actually look like—and why is the industry moving toward centralized computing and zonal architectures?

In this podcast, Sally discusses how AI is lowering the barriers to finding security vulnerabilities, and how hardware could be the next frontier, with Arteris’ Andreas Kuehlmann. 

MIPS CEO Sameer Wasson explains why AI workloads should drive silicon design, and how MIPS, ARC and GlobalFoundries aim to lower the barriers to custom AI hardware.

On Axelera's fifth anniversary, CEO Fabrizio del Maffeo discusses funding, talent, RISC-V, edge AI and why Europe needs to think bigger.

In this podcast, Tensordyne’s Gilles Backhus explains why the company abandoned conventional arithmetic in pursuit of dramatically more efficient AI inference with the logarithmic number system.