One-transistor synapse cell with weight adjustment

    公开(公告)号:US10395713B2

    公开(公告)日:2019-08-27

    申请号:US15859583

    申请日:2017-12-31

    IPC分类号: G06N3/06 G11C11/22 H03K19/177

    摘要: Word lines intersect bit lines at a plurality of cross points where a plurality of single memory transistor synapse cells are located. Each cell includes a memory transistor; a pulse shaping unit coupled to a given one of a plurality of signal lines and a gate of the memory transistor; a logic gate having inputs coupled to a corresponding one of the word lines and a corresponding one of the bit lines, and an output coupled to the pulse shaping unit; and a pass gate arrangement. The latter is coupled to the memory transistor, the corresponding one of the word lines, the corresponding one of the bit lines, and the output of the logic gate. Pulses are applied to the gate of the memory transistor for weight adjustment during update and to interconnect the memory transistor to the corresponding one of the bit lines during inference.

    Mixed precision capable hardware for tuning a machine learning model

    公开(公告)号:US11604647B2

    公开(公告)日:2023-03-14

    申请号:US16558536

    申请日:2019-09-03

    摘要: An apparatus includes a memory and a processor coupled to the memory. The processor includes first and second sets of arithmetic units having first and second precision for floating-point computations, the second precision being lower than the first precision. The processor is configured to obtain a machine learning model trained in the first precision, to utilize the second set of arithmetic units to perform inference on input data, to utilize the first set of arithmetic units to generate feedback for updating parameters of the second set of arithmetic units based on the inference performed on the input data by the second set of arithmetic units, to tune parameters of the second set of arithmetic units based at least in part on the feedback generated by the first set of arithmetic units, and to utilize the second set of arithmetic units with the tuned parameters to generate inference results.

    Multi-kernel configuration for convolutional neural networks

    公开(公告)号:US11556763B2

    公开(公告)日:2023-01-17

    申请号:US16279416

    申请日:2019-02-19

    摘要: Methods and systems of implementing a convolutional neural network are described. In an example, a structure may receive input signals and distribute the input signals to a plurality of unit cells. The structure may include a plurality of multi-kernel modules that may include a respective set of unit cells. A unit cell may correspond to an element of a kernel being implemented in the convolutional neural network and may include a storage component configured to store a weight of a corresponding element of the kernel. A first pass gate of the unit cell may be activated to pass a stored weight of the unit cell to a plurality of operation circuits in the corresponding unit cell, such that the stored weight may be applied to the input signals. The structure may generate a set of outputs based on the application of the stored weights to the input signals.

    MACHINE LEARNING HARDWARE HAVING REDUCED PRECISION PARAMETER COMPONENTS FOR EFFICIENT PARAMETER UPDATE

    公开(公告)号:US20210064985A1

    公开(公告)日:2021-03-04

    申请号:US16558585

    申请日:2019-09-03

    IPC分类号: G06N3/08 G06N3/04

    摘要: An apparatus for training and inferencing a neural network includes circuitry that is configured to generate a first weight having a first format including a first number of bits based at least in part on a second weight having a second format including a second number of bits and a residual having a third format including a third number of bits. The second number of bits and the third number of bits are each less than the first number of bits. The circuitry is further configured to update the second weight based at least in part on the first weight and to update the residual based at least in part on the updated second weight and the first weight. The circuitry is further configured to update the first weight based at least in part on the updated second weight and the updated residual.

    FET based synapse network
    16.
    发明授权

    公开(公告)号:US10892330B2

    公开(公告)日:2021-01-12

    申请号:US15202729

    申请日:2016-07-06

    发明人: Jin P. Han Xiao Sun

    摘要: A synapse network device includes an array of field effect transistor (FET) devices having controllable channel resistance. Pre-neurons are coupled to the array to provide input pulses to the array on first terminals of the FET devices. Post-neurons are coupled to the array to receive outputs from the array on second terminals of the FET devices and provide feedback to the array on third terminals of the FET devices, wherein a state of the FET devices is indicated based upon signals applied to the FET devices.

    Ferro-electric complementary FET
    18.
    发明授权

    公开(公告)号:US10658384B2

    公开(公告)日:2020-05-19

    申请号:US16576441

    申请日:2019-09-19

    摘要: A field-effect transistor includes a semiconductor substrate having first, second, third, and fourth sides, and a ferroelectric gate stack on an upper surface of the substrate. The ferroelectric gate stack includes a gate insulating layer; and a ferroelectric material layer on the gate insulating layer. Portions of the upper surface of the substrate between the first side and the ferroelectric gate stack and between the second side and the ferroelectric gate stack are doped with n-type impurities, and portions of the upper surface of the substrate between the third side and the ferroelectric gate stack and between the fourth side and the ferroelectric gate stack are doped with p-type impurities. A presence of both n and p channels in a same region increases a capacitance and voltage gain of the ferroelectric gate stack.

    CIRCUITRY FOR ONE-TRANSISTOR SYNAPSE CELL AND OPERATION METHOD OF THE SAME

    公开(公告)号:US20190378555A1

    公开(公告)日:2019-12-12

    申请号:US16550809

    申请日:2019-08-26

    IPC分类号: G11C11/22 H03K19/177

    摘要: Word lines intersect bit lines at a plurality of cross points where a plurality of single memory transistor synapse cells are located. Each cell includes a memory transistor; a pulse shaping unit coupled to a given one of a plurality of signal lines and a gate of the memory transistor; a logic gate having inputs coupled to a corresponding one of the word lines and a corresponding one of the bit lines, and an output coupled to the pulse shaping unit; and a pass gate arrangement. The latter is coupled to the memory transistor, the corresponding one of the word lines, the corresponding one of the bit lines, and the output of the logic gate. Pulses are applied to the gate of the memory transistor for weight adjustment during update and to interconnect the memory transistor to the corresponding one of the bit lines during inference.