Columbia University

Arizona State University

Marvell Asia Pte., Ltd.

Nokia Technologies Oy

Zhewei Jiang

Jae-sun Seo

Mingoo Seok

Shihui Yin

Jack M Higman

Muhammed Ahosan Ul Karim

Xi Cao

Vivek Joshi

Hungkei Chow

In some embodiments, an in-memory-computing SRAM macro based on capacitive-coupling computing (C3) (which is referred to herein as 'C3SRAM') is provided. In some embodiments, a C3SRAM macro can support array-level fully parallel computation, multi-bit outputs, and configurable multi-bit inputs. The macro can include circuits embedded in bitcells and peripherals to perform hardware acceleration for neural networks with binarized weights and activations in some embodiments. In some embodiments, the macro utilizes analog-mixed-signal capacitive-coupling computing to evaluate the main computations of binary neural networks, binary-multiply-and-accumulate operations. Without needing to access the stored weights by individual row, the macro can assert all of its rows simultaneously and form an analog voltage at the read bitline node through capacitive voltage division, in some embodiments. With one analog-to-digital converter (ADC) per column, the macro cab realize fully parallel vector-matrix multiplication in a single cycle in accordance with some embodiments.

Circuits and methods for in-memory computing

A static random-access memory (SRAM) system includes SRAM cells configured to perform exclusive NOR operations between a stored binary weight value and a provided binary input value. In some embodiments, SRAM cells are configured to perform exclusive NOR operations between a stored binary weight value and a provided ternary input value. The SRAM cells are suitable for the efficient implementation of emerging deep neural network technologies such as binary neural networks and XNOR neural networks.

Static random-access memory for deep neural networks

Disclosed is a three-port static random access memory (3P-SRAM) that performs XNOR operations. The cell has a write port and first and second read ports. Read operations are enabled through either the first read port using a first read wordline and a common read bitline or the second read port using a second read wordline and the common read bitline. Read wordline activation is controlled such that only one read wordline is activated (i.e., receives a read pulse) at a time. As a result, a read operation through either read port effectively accomplishes an XNOR operation. Also disclosed is a memory array, which incorporates such cells and which performs XNOR-bitcount-compare functions. Since XNOR-bitcount-compare functions are used in XNOR-NET type binary neural networks (BNNs), the memory array can be employed for implementing such a BNN designed for improved performance, scalability, and manufacturability. Also disclosed is an in-memory computing method.

Three-port memory cell and array for in-memory computing

Growing community of inventors

Oakland Gardens, NY, United States of America

Zhewei Jiang