Advanced Micro Devices Corporation

Vignesh Adhinarayanan

Jagadish B Kotra

Sriseshan Srikanth

Sergey Blagodurov

Michael Ignatowski

Shomit N Das

Anthony T Gutierrez

Vedula Venkata Srikant Bharadwaj

Hyung-Dong Lee

Shaizeen Dilawarhusen Aga

Ashwin Aji

Ioannis Papadopoulos

A fine-grained dynamic random-access memory (DRAM) includes a first memory bank, a second memory bank, and a dual mode I/O circuit. The first memory bank includes a memory array divided into a plurality of grains, each grain including a row buffer and input/output (I/O) circuitry. The dual-mode I/O circuit is coupled to the I/O circuitry of each grain in the first memory bank, and operates in a first mode in which commands having a first data width are routed to and fulfilled individually at each grain, and a second mode in which commands having a second data width different from the first data width are fulfilled by at least two of the grains in parallel.

Method and apparatus for recovering regular access performance in fine-grained DRAM

A system includes memory hardware including a memory and a processing-in-memory component. A system includes a host including at least one core. A system includes a memory controller including a scheduling system. The scheduling system transforms an all-bank processing-in-memory command into multiple masked processing-in-memory commands. The scheduling system also schedules the multiple masked processing-in-memory commands to the processing-in-memory component.

Transforming all-bank processing-in-memory operations into multiple masked processing-in-memory operations

An apparatus and method for creating less computationally intensive nodes for a neural network. An integrated circuit includes a host processor and multiple memory channels, each with multiple memory array banks. Each of the memory array banks includes components of a processing-in-memory (PIM) accelerator and a scatter and gather circuit used to dynamically perform quantization operations and dequantization operations that offload these operations from the host processor. The host processor executes a data model that represents a neural network. The memory array banks store a single copy of a particular data value in a single precision. Therefore, the memory array banks avoid storing replications of the same data value with different precisions to be used by a neural network node. The memory array banks dynamically perform quantization operations and dequantization operations on one or more of the weight values, input data values, and activation output values of the neural network.

Method and apparatus for quantization and dequantization of neural network input and output data using processing-in-memory

A method includes, in response to receiving a command from a processing device, reading original data from a set of one or more memory devices based on an address range specified in the command, and transmitting a subset of the original data to the processing device, where the subset includes fewer zero values than the original data.

Near-memory engine for reducing bandwidth utilization in sparse data applications

Systems, methods, devices, and computer-implemented instructions for processor power management implemented in a compiler. In some implementations, a characteristic of code is determined. An instruction based on the determined characteristic is inserted into the code. The code and inserted instruction are compiled to generate compiled code. The compiled code is output.

Compiler directed fine grained power management

A memory sprint controller, responsive to an indicator of an irregular memory access phase, causes a memory controller to enter a sprint mode in which it temporarily adjusts at least one timing parameter of a dynamic random access memory (DRAM) to reduce a time in which a designated number of activate (ACT) commands are allowed to be dispatched to the DRAM.

Growing community of inventors

Austin, TX, United States of America

Vignesh Adhinarayanan