Gyrfalcon Technology Inc.

Magic Technologies, Inc.

Headway Technologies, Incorporated

Daniel H Liu

Chyu-Jiuh Torng

Tom Zhong

Wai-Ming Johnson Kan

Adam Zhong

Rongfu Xiao

Lin Yang

Qi Dong

Wenhan Zhang

Hualiang Yu

This disclosure relates to testing of integrated artificial intelligence (AI) circuit with embedded memory to improve effective chip yield and to mapping addressable memory segments of the embedded memory to multilayer AI networks at the network level, layer level, parameter level, and bit level based on bit error rate (BER) of the addressable memory segments. The disclosed methods and systems allows for deployment of one or more multilayer AI networks in an AI circuit with sufficient model accuracy even when the embedded memory has an overall BER higher than a preferred overall threshold.

Effective chip yield for artificial intelligence integrated circuit with embedded memory

An integrated circuit includes an artificial intelligence (AI) logic and an embedded memory coupled to the AI logic and connectable to an external processor. The embedded memory includes multiple storage cells and multiple reference units. One or more reference units in the memory are selected for memory access through configuration at chip packaging level by the external processor. The external processor may execute a self-test process to select or update the one or more reference units for memory access so that the error rate of memory is below a threshold. The self-test process may be performed, via a memory initialization controller in the memory, to test and reuse the reference cells in the memory at chip level. The embedded memory may be a STT-MRAM, SOT, OST MRAM, and/or MeRAM memory.

Method of self-testing and reusing of reference cells in a memory architecture

CNN (Cellular Neural Networks or Cellular Nonlinear Networks) based digital Integrated Circuit for artificial intelligence contains multiple CNN processing units. Each CNN processing unit contains CNN logic circuits operatively coupling to a memory subsystem having first and second memories. The first memory contains magnetic random access memory (MRAM) cells for storing weights (e.g., filter coefficients) while the second memory is for storing input signals (e.g., imagery data). The first memory may store one-time-programming weights. The memory subsystem may contain a third memory that contains MRAM cells for storing one-time-programming data for security purpose. The second memory contains MRAM cells or static random access memory cells. Each MRAM cell contains a Spin-Orbit-Torque (SOT) based magnetic tunnel junction (MTJ) element.

Memory subsystem in CNN based digital IC for artificial intelligence

CNN (Cellular Neural Networks or Cellular Nonlinear Networks) based digital Integrated Circuit for artificial intelligence contains multiple CNN processing units. Each CNN processing unit contains CNN logic circuits operatively coupling to a memory subsystem having first and second memories. The first memory contains magnetic random access memory (MRAM) cells for storing weights (e.g., filter coefficients) while the second memory is for storing input signals (e.g., imagery data). The first memory may store one-time-programming weights or filter coefficients. The memory subsystem may contain a third memory that contains MRAM cells for storing one-time-programming data for security purpose. The second memory contains MRAM cells or static random access memory cells. Each MRAM cell contains a voltage-controlled magnetic anisotropy (VCMA) based magnetic tunnel junction (MTJ) element. Magnetization direction in VCMA based MTJ element can be in-plane or out-of-plane.

CMOS devices are provided in a substrate having a topmost metal layer comprising metal landing pads and metal connecting pads. A plurality of magnetic tunnel junction (MTJ) structures are provided over the CMOS devices and connected to the metal landing pads. The MTJ structures are covered with a dielectric layer that is polished until the MTJ structures are exposed. Openings are etched in the dielectric layer to the metal connecting pads. A seed layer is deposited over the dielectric layer and on inside walls and bottom of the openings. A copper layer is plated on the seed layer until the copper layer fills the openings. The copper layer is etched back and the seed layer is removed. Thereafter, an aluminum layer is deposited over the dielectric layer, contacting both the copper layer and the MTJ structures, and patterned to form a bit line.

Method of spin torque MRAM process integration

A STT-MRAM integration scheme is disclosed wherein the connection between a MTJ and CMOS metal is simplified by forming an intermediate via contact (VAC) on a CMOS landing pad, a metal (VAM) pad that contacts and covers the VAC, and a MTJ on the VAM. A dual damascene process is performed to connect BIT line metal to CMOS landing pads through VAC/VAM/MTJ stacks in a device region, and to connect BIT line connection pads to CMOS connection pads through BIT connection vias outside the device region. The VAM pad is a single layer or composite made of Ta, TaN, or other conductors which serves as a diffusion barrier, has a highly smooth surface for MTJ formation, and provides excellent selectivity with refill dielectric materials during a chemical mechanical polish process. Each VAC is from 500 to 3000 Angstroms thick to minimize additional circuit resistance and minimize etch burden.

High density spin-transfer torque MRAM process

Described herein are novel, cost effective and scalable methods for integrating a CMOS level with a memory cell level to form a field induced MRAM device. The memory portion of the device includes N parallel word lines, which may be clad, overlaid by M parallel bit lines orthogonal to the word lines and individual patterned memory cells formed on previously patterned electrodes at the N&#xd7;M intersections of the two sets of lines. The memory portion is integrated with a CMOS level and the connection between levels is facilitated by the formation of interconnecting vias between the N&#xd7;M electrodes and corresponding pads in the CMOS level and by word line connection pads in the memory device level and corresponding metal pads in the CMOS level. Of particular importance are process steps that replace single damascene formations by dual damascene formations, different process steps for the formation of clad and unclad word lines and the formation of patterned electrodes for the memory cells prior to the patterning of the cells themselves.

Growing community of inventors

San Jose, CA, United States of America

Daniel H Liu