Feds Aid Chip Law; Garage Fab Hits ICs
Summary
Advancements in both industrial scaling initiatives and grassroots fabrication signal a dynamic, multi-pronged approach to overcoming current hardware limitations.
- Gelsinger’s New Venture Pat Gelsinger is now leading xLight, seeking federal aid to sustain semiconductor scaling efforts past current limits 1.
- AI Hardware Evolution Google’s TPU development, initiated in 2013, focused on scaling neural network demands beyond standard hardware capabilities 2.
- Efficiency Breakthrough Zebra-Llama introduces hybrid models that achieve extreme efficiency, potentially reducing required compute by up to 60% 4.
- Garage Fabrication Success A college senior successfully fabricated a second integrated circuit (Z2) using lithography in a home setup, replicating earlier efforts 5.
- 2013 - Year Google initiated the development of the specialized Tensor Processing Unit (TPU) 2.
- 6 - The number of transistors in the first Z1 amplifier fabricated in the garage fab in 2018 5.
- 60% - Potential reduction in compute achieved by the highly efficient Zebra-Llama hybrid model architecture 4.
Key Moments
-
Pat Gelsinger, one year after leaving Intel, is now a general partner at Playground Global and executive chairman for the semiconductor startup xLight
— Article [1] -
The paper 'Zebra-Llama: Towards Extremely Efficient Hybrid Models,' introduces a method for hybrid model architectures.
— Article [4] -
In 2018, the author first fabricated integrated circuits in a garage fab, creating the Z1 amplifier with 6 transistors.
— Article [5] -
Google’s development of the Tensor Processing Unit (TPU) was initiated in 2013 after the company realized scaling neural network demand
— Article [2]
Different Perspectives
Supporting View
The home-built Oblast project demonstrates sustained interest and capability in retro-hardware development for platforms like the Commodore 64.
Sources:
[3]