Print this page

Reinforcement learning (RL) algorithm shortens human time

Saturday, 12 June 2021
Reduce the time of designing specialized chips for tensioner processing units Reduce the time of designing specialized chips for tensioner processing units

Vision, Dubai United Arab Emirates:- Published in Nature, a new study says that a Reinforcement Learning (RL) algorithm has been able to design specialized chips for tensor processing units - a type of chip used in artificial intelligence - in record time compared to humans.

While it takes the human brain months to design these chips, the reinforcement learning (RL) algorithm is able to manufacture them in just six hours.

The researchers wrote in the study: “The RL agent is getting better and faster at optimizing floorplans because it generates a greater number of network lists. It appears that it can generate chip floor plans comparable to or superior to human experts in less than six hours, while it takes humans months to produce floor plans. acceptable for modern accelerators.

Google researchers gave the program 10,000 floor plans to slice it for analysis, and then figured out how to come up with floor plans that don't use more space, wires, and electrical power than those designed by humans.

The chip floor plan is where parts such as CPUs, GPUs, and memory are placed on the silicon.

And since the 1960s, there have been three different approaches to how these parts can be placed on silicon: segmentation-based methods, a stochastic approach, and analytics.

None of them have achieved the level of human performance, but the RL system is able to do so fairly easily.

Rate this item
(0 votes)