Bio:

Sven Beyer received his master’s and Ph.D. in Physics from the University of Hamburg, Germany. He started his career with Infineon as a manufacturing engineer in the etch department in 2003. Joining the integration department of AMD in 2005, he intermediately spent a year in the IBM ASTA alliance 2007 working on the 45nm node. Returning to Dresden, Sven has worked in many roles since then, ranging from integration technology-lead to customer engineering and lasting throughout the separation of GlobalFoundries and AMD. Today he serves as DMTS in GlobalFoundries FAB1 as technology architect (TA), overseeing mainly the eNVM & CMOS roadmap and development in Dresden.

Abstract:

Introducing the “AI stack” at the example of transformer-based AI and laying out current and potential future ownership of the elements of this stack, I will make the daring claim that “Attention ain’t everything”. Hence, pushing the digital brute force computing capability of leading-edge CMOS alone, by continuing to stumble along the path of Moore’s law, will likely never deliver true human like intelligence. Perusing this way, striving for true human like intelligence, power consumption will sky-rocket, eventually building an insurmountable barrier. It will need another paradigm shift, already foreseeable at the horizon, to lift AI capabilities on a manageable power budget, to the next level. Here the value-added solution (VAS) device offerings of mature CMOS nodes, like non-volatile memory cells, in combination with specialized low power CMOS platforms, might play a crucial role. In addition, financially affordable CMOS technologies bear the chance to bring this fascinating technology into the edge devices.