Google Unveils AlphaChip: An AI for Designing Processors, Already Adopted by MediaTek
Google has introduced AlphaChip, a neural network designed to create chip layouts. This AI aims to significantly accelerate the processor design process, optimizing performance, power consumption, and chip area. The method has already been tested in developing Google's Tensor processors and adopted by other companies, including MediaTek.
Processor layout is considered the most time-consuming and labor-intensive phase of chip development. While Synopsys has created an AI to assist humans in this area, which Samsung has tested, Google is also focused on simplifying the process. Traditionally, it takes engineers around 24 months to design a GPU within a chipset and several months for simpler elements, leading to significant expenses for companies. Google claims AlphaChip can dramatically accelerate this process, generating a chip layout in just a few hours. Moreover, the results are said to be "excellent," as the AI optimizes both energy efficiency and performance.
AlphaChip is based on a reinforcement learning model where an agent takes actions in a given environment to make better choices in the future. The AI treats chip planning as a game, placing one circuit component at a time on an empty grid.
Since 2020, AlphaChip has been used to develop Google's custom AI accelerators powering the company's cloud services and Gemini chatbot, as well as MediaTek's Dimensity 5G chips. Currently, both Google and MediaTek rely on AlphaChip for a limited set of blocks, with humans still handling the majority of the design work. However, Google is exploring how AlphaChip can be utilized in later stages of chip development.
