Killing asymmetric resistive processing units for neural network training

This invention describes a specialized electronic circuit called a resistive processing unit (RPU) array, which is designed to help train neural networks more efficiently. Before the training begins, the method identifies individual RPUs within the array that are faulty because they don't behave consistently (are "asymmetric"). These problematic RPUs are then permanently disabled, or "burned," by applying a high electrical voltage, ensuring only reliable units are used for training.

Why it matters: Filed before the explosion of large-scale neural networks, the need for energy-efficient, reliable AI hardware has become paramount since 2016. This invention addresses a critical manufacturing and operational reliability challenge for RPU arrays, which are increasingly vital for advanced AI training.

Status: ExpiredProtection ended and was never renewed
How hard to build: SpecializedCustom silicon fabrication, precise electrical control

3 AI reinvention directions for this patent

See all three

AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.

Reinvent this with AI

Where this one came from

Every morning we go through millions of abandoned patents and pick out 15 worth your time. This was one of today’s.

This one

+10