MUMBAI, India, Sept. 7 -- Intellectual Property India has published a patent application (202641101979 A) filed by Centurion University Of Technology And Management Cutm, Odisha; and Centurion University Of Technology And Management Cutm, Andhra Pradesh on August 24, 2026, for Hierarchical Urdhva-Tiryagbhyam Vedic Multiply-Accumulate Engine For Int4 Neural-Network Inference Acceleration.
Inventors include Satyanarayan Padhy; Chandra Sekhar Dash; and Saswat Panda.
The application for the patent was published on September 04, 2026, under issue no. 36/2026.
Abstract: The present disclosure relates to a hierarchical Urdhva-Tiryagbhyam Vedic multiply-accumulate engine (100) for processing INT4-quantized neural-network operands. The engine (100) includes an operand input interface (102) adapted to receive first and second 4-bit operands, and an operand splitting module (104) adapted to divide each operand into higher-order and lower-order 2-bit portions. Four structurally identical reusable 2×2 Vedic multiplier sub-tiles (106a–106d) concurrently generate four corresponding partial products using combinational partial-product and crosswise addition logic. A bit-weight alignment module (110) positionally aligns the partial products by zero-, two-, two-, and four-bit shifts, respectively. A partial-product addition module (112) combines the aligned partial products to generate an 8-bit product, which is provided through a product output interface (114). An accumulator module (116) adds the 8-bit product to a previously stored accumulated value, thereby enabling single-interval multiplication and successive accumulation for neural-network inference operations. Fig. 1 is the reference figure.
Disclaimer: Curated by HT Syndication.
