公开(公告)号：IL297848A

公开(公告)日：2023-01-01

申请号：IL29784822

申请日：2022-10-31

Applicant: IBM CORP ,  HSINYU TSAI ,  SANJAY KARIYAPPA

Inventor： HSINYU TSAI ,  SANJAY KARIYAPPA

IPC: G06V10/75

Abstract: Drift regularization is provided to counteract variation in drift coefficients in analog neural networks. In various embodiments, a method of training an artificial neural network is illustrated. A plurality of weights is randomly initialized. Each of the plurality of weights corresponds to a synapse of an artificial neural network. At least one array of inputs is inputted to the artificial neural network. At least one array of outputs is determined by the artificial neural network based on the at least one array of inputs and the plurality of weights. The at least one array of outputs is compared to ground truth data to determine a first loss. A second loss is determined by adding a drift regularization to the first loss. The drift regularization is positively correlated to variance of the at least one array of outputs. The plurality of weights is updated based on the second loss by backpropagation.

公开(公告)号：IL297331A

公开(公告)日：2022-12-01

申请号：IL29733122

申请日：2022-10-13

Applicant: IBM ,  HSINYU TSAI ,  GEOFFREY BURR ,  PRITISH NARAYANAN

Inventor： HSINYU TSAI ,  GEOFFREY BURR ,  PRITISH NARAYANAN

Abstract: Implementing a convolutional neural network (CNN) includes configuring a crosspoint array to implement a convolution layer in the CNN. Convolution kernels of the layer are stored in crosspoint devices of the array. Computations for the CNN are performed by iterating a set of operations for a predetermined number of times. The operations include transmitting voltage pulses corresponding to a subpart of a vector of input data to the crosspoint array. The voltage pulses generate electric currents that are representative of performing multiplication operations at the crosspoint device based on weight values stored at the crosspoint devices. A set of integrators accumulates an electric charge based on the output electric currents from the respective crosspoint devices. The crosspoint array outputs the accumulated charge after iterating for the predetermined number of times. The accumulated charge represents a multiply-add result of the vector of input data and the one or more convolution kernels.

公开(公告)号：GB2547121A

公开(公告)日：2017-08-09

申请号：GB201704200

申请日：2016-02-08

Applicant: IBM

Inventor： JOY CHENG ,  MELIA TJIO ,  GURPREET SINGH ,  ALEXANDER FRIZ ,  HSINYU TSAI ,  MARKUS BRINK ,  MICHAEL GUILLORN ,  CHI-CHUN LIU ,  GREGORY DOERK ,  DANIEL PAUL SANDERS

IPC: C08J7/04 ,  H01L21/02 ,  H01L21/311 ,  H01L29/06 ,  H05K3/00

Abstract: Hybrid pre-patterns were prepared for directed self-assembly of a given block copolymer capable of forming a lamellar domain pattern. The hybrid pre-patterns have top surfaces comprising independent elevated surfaces interspersed with adjacent recessed surfaces. The elevated surfaces are neutral wetting to the domains formed by self-assembly. Material below the elevated surfaces has greater etch-resistance than material below the recessed surfaces in a given etch process. Following other dimensional constraints of the hybrid pre-pattern described herein, a layer of the given block copolymer was formed on the hybrid pre-pattern. Self- assembly of the layer produced a lamellar domain pattern comprising self-aligned, unidirectional, perpendicularly oriented lamellae over the elevated surfaces, and parallel and/or perpendicularly oriented lamellae over recessed surfaces. The domain patterns displayed long range order along the major axis of the pre-pattern. The lamellar domain patterns are useful in forming transfer patterns comprising two-dimensional customized features.

公开(公告)号：IL297848B1

公开(公告)日：2025-04-01

申请号：IL29784822

申请日：2022-10-31

Applicant: IBM ,  HSINYU TSAI ,  SANJAY KARIYAPPA

Inventor： HSINYU TSAI ,  SANJAY KARIYAPPA

IPC: G06V10/75

Abstract: Drift regularization is provided to counteract variation in drift coefficients in analog neural networks. In various embodiments, a method of training an artificial neural network is illustrated. A plurality of weights is randomly initialized. Each of the plurality of weights corresponds to a synapse of an artificial neural network. At least one array of inputs is inputted to the artificial neural network. At least one array of outputs is determined by the artificial neural network based on the at least one array of inputs and the plurality of weights. The at least one array of outputs is compared to ground truth data to determine a first loss. A second loss is determined by adding a drift regularization to the first loss. The drift regularization is positively correlated to variance of the at least one array of outputs. The plurality of weights is updated based on the second loss by backpropagation.

公开(公告)号：GB2547121B

公开(公告)日：2019-09-18

申请号：GB201704200

申请日：2016-02-08

Applicant: IBM

Inventor： JOY CHENG ,  MELIA TJIO ,  GURPREET SINGH ,  ALEXANDER FRIZ ,  HSINYU TSAI ,  MARKUS BRINK ,  MICHAEL GUILLORN ,  CHI-CHUN LIU ,  GREGORY DOERK ,  DANIEL PAUL SANDERS

IPC: C08J7/04 ,  B05D1/18 ,  G03F7/00 ,  H01L21/02 ,  H01L21/033 ,  H01L21/311 ,  H01L29/06 ,  H05K3/00

Abstract: Hybrid pre-patterns were prepared for directed self-assembly of a given block copolymer capable of forming a lamellar domain pattern. The hybrid pre-patterns have top surfaces comprising independent elevated surfaces interspersed with adjacent recessed surfaces. The elevated surfaces are neutral wetting to the domains formed by self-assembly. Material below the elevated surfaces has greater etch-resistance than material below the recessed surfaces in a given etch process. Following other dimensional constraints of the hybrid pre-pattern described herein, a layer of the given block copolymer was formed on the hybrid pre-pattern. Self-assembly of the layer produced a lamellar domain pattern comprising self-aligned, unidirectional, perpendicularly oriented lamellae over the elevated surfaces, and parallel and/or perpendicularly oriented lamellae over recessed surfaces. The domain patterns displayed long range order along the major axis of the pre-pattern. The lamellar domain patterns are useful in forming transfer patterns comprising two-dimensional customized features.