Amazon patents chip design that simplifies AI math steps
Amazon describes a hardware circuit for running AI models that combines several mathematical adjustments into a single step. This design merges data scaling, normalization, and encoding tasks to prepare numbers for the next stage of the AI's calculation.
In the filing’s own words
Some embodiments provide a neural network inference circuit (NNIC). The NNIC includes a set of core computation circuits for executing a core computation on a set of input values for a neural network computation node to generate a first value. The NNIC includes a set of affine transform circuits for applying an affine transform to the first value to generate a second value. The applied affine transform is a combination of (i) a batch normalization transform, (ii) a learned affine transform, (iii) a weight scale for the computation node, and (iv) an encoding transform such that the second value is encoded for input to an activation function. The NNIC includes a set of activation circuits for applying the activation function to the second value to generate an output value for the computation node.
What exactly they claim (claim 1)
1. A neural network inference circuit comprising: a set of core computation circuits configured to: generate, using a set of input values for a neural network, a first value, wherein the first value is output of a core computation for a first node of the neural network; a set of affine transform circuits configured to: generate, based on the first value, a second value, wherein generating the first value comprises: applying a first affine transform to the second value, wherein the first affine transform is a combination of (i) a batch normalization transform, (ii) a learned affine transform, (iii) a weight scale for the first node, and (iv) an encoding transform associated with a first activation function such that the second value is encoded for input in a first input range associated with the first activation function; and a set of activation circuits configured to apply the first activation function to the second value to generate an output value for the first node.
Facts
- Company
- Amazon (Amazon Technologies, Inc.)
- Official title
- Neural network circuit using affine transforms to encode activations
- Patent number
- US12762389B1
- Application number
- 18115580
- Filed
- Feb 28, 2023
- Granted
- Oct 13, 2026
- Inventors
- 2
- Claims
- 20
Classified under
G06N 3/063 Physics & computingG06N 3/048 Physics & computingMore from Amazon in Chips & hardware
- Amazon patents using metal frames to send data between devices Sep 1, 2026
- Amazon patents flexible mat that detects when items are placed or removed Sep 15, 2026
- Amazon patents hardware chip that watches memory for security threats Sep 29, 2026
- Amazon patents modular radio chips that swap wireless standards Oct 20, 2026
- Amazon patents tool to auto-update chip test scripts Sep 1, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.