Industry Expert Blogs

In this AI-driven era of pervasive intelligence, our silicon and systems customers face unprecedented pressure to deliver the increasing compute performance required to train LLM-based AI systems as demands double every six months. Moreover, they’re being challenged to achieve sustainable computing - exponential performance gains with increasing power efficiency. Traditional reliance on Moore's Law is no longer sufficient, as recent node transitions no longer consistently deliver the expected 2X improvement in performance, power, and area.

These challenges are compounded by an expected semiconductor workforce shortage and increasing design complexity as we march towards trillion-transistor systems by the end of this decade. And yet remarkably – contrary to these trends – the pace of semiconductor innovation is accelerating.

Just look at recent announcements from AMD and NVIDIA at Computex. These major chip makers not only showcased new AI processors featuring hundreds of billions of transistors and faster, denser memory, destined for leading-edge manufacturing nodes, they also put a spotlight on their increasing speed of innovation. Despite mounting complexity, product refresh cycles for new AI processors are contracting from 18-24 months to 12 months.

Click here to read more ...