Method and apparatus for defect-tolerant memory-based artificial neural network

    公开(公告)号:US12205017B2

    公开(公告)日:2025-01-21

    申请号:US18231769

    申请日:2023-08-08

    Abstract: Disclosed is a methods and apparatus which can improve defect tolerability of a hardware-based neural network. In one embodiment, a method for performing a calculation of values on first neurons of a first layer in a neural network, includes: receiving a first pattern of a memory cell array; determining a second pattern of the memory cell array according to a third pattern; determining at least one pair of columns of the memory cell array according to the first pattern and the second pattern; switching input data of two columns of each of the at least one pair of columns of the memory cell array; and switching output data of the two columns in each of the at least one pair of columns of the memory cell array so as to determine the values on the first neurons of the first layer.

    LOW-DROPOUT REGULATOR CIRCUIT
    2.
    发明公开

    公开(公告)号:US20230393598A1

    公开(公告)日:2023-12-07

    申请号:US18232772

    申请日:2023-08-10

    CPC classification number: G05F1/56

    Abstract: A voltage regulation circuit includes a voltage regulator that is configured to provide a stable output voltage based on an input voltage; and a control circuit, coupled to the voltage regulator, and configured to provide an injection current to maintain the stable output voltage in response to an enable signal provided at an input of the control circuit transitioning to a predetermined state and cease providing the injection current when the control circuit detects that a voltage level of the output voltage is higher than a pre-defined voltage level.

    Method and apparatus for defect-tolerant memory-based artificial neural network

    公开(公告)号:US11461623B2

    公开(公告)日:2022-10-04

    申请号:US16542049

    申请日:2019-08-15

    Abstract: Disclosed is a methods and apparatus which can improve defect tolerability of a hardware-based neural network. In one embodiment, a method for performing a calculation of values on first neurons of a first layer in a neural network, includes: receiving a first pattern of a memory cell array; determining a second pattern of the memory cell array according to a third pattern; determining at least one pair of columns of the memory cell array according to the first pattern and the second pattern; switching input data of two columns of each of the at least one pair of columns of the memory cell array; and switching output data of the two columns in each of the at least one pair of columns of the memory cell array so as to determine the values on the first neurons of the first layer.

    MTJ-based temperature-sensing device

    公开(公告)号:US10288494B2

    公开(公告)日:2019-05-14

    申请号:US15591976

    申请日:2017-05-10

    Abstract: A thermometer circuit configured to estimate a monitored temperature is disclosed. The circuit includes an adjustable resistor presenting a first resistance value that is temperature-independent and a second resistance value that is temperature-dependent, wherein a first current signal is conducted across the resistor when it presents the first resistance value and a second current signal is conducted across the resistor when it presents the second resistance value; a plurality of gated conductors coupled to the resistor; and a control circuit, coupled to the resistor and the plurality of gated conductors, and configured to selectively deactivate at least one of the plurality of gated conductors to compare the first and second current signals to estimate the monitored temperature.

    Circuits and methods for limiting current in random access memory cells

    公开(公告)号:US09620209B2

    公开(公告)日:2017-04-11

    申请号:US15238167

    申请日:2016-08-16

    Abstract: Circuits and methods for limiting cell current or throttling write operation, or both, in resistive random access memory (RRAM or ReRAM) cells are provided. An RRAM cell can include a select transistor and a programmable resistor that can change between a relatively high resistance and a relatively low resistance. The present circuits and methods can reduce or inhibit excess current from being applied to the programmable resistor, which potentially can regulate the resistance of the programmable resistor so as to reduce or inhibit decreases in the resistance of that resistor below the relatively low resistance. Such regulation potentially can improve reliability of the RRAM cell. Additionally, or alternatively, the present circuits and methods can throttle a write operation in an RRAM cell, e.g., can disable current flow through the RRAM cell based on the programmable resistor reaching a pre-defined target resistance, such as the relatively low resistance.

    Low-dropout voltage regulation circuit

    公开(公告)号:US11797034B2

    公开(公告)日:2023-10-24

    申请号:US17339818

    申请日:2021-06-04

    CPC classification number: G05F1/56

    Abstract: A voltage regulation circuit includes a voltage regulator that is configured to provide a stable output voltage based on an input voltage; and a control circuit, coupled to the voltage regulator, and configured to provide an injection current to maintain the stable output voltage in response to an enable signal provided at an input of the control circuit transitioning to a predetermined state and cease providing the injection current when the control circuit detects that a voltage level of the output voltage is higher than a pre-defined voltage level.

    METHOD AND APPARATUS FOR DEFECT-TOLERANT MEMORY BASED ARTIFICIAL NEURAL NETWORK

    公开(公告)号:US20220383085A1

    公开(公告)日:2022-12-01

    申请号:US17883594

    申请日:2022-08-08

    Abstract: Disclosed is a methods and apparatus which can improve defect tolerability of a hardware-based neural network. In one embodiment, a method for performing a calculation of values on first neurons of a first layer in a neural network, includes: receiving a first pattern of a memory cell array; determining a second pattern of the memory cell array according to a third pattern; determining at least one pair of columns of the memory cell array according to the first pattern and the second pattern; switching input data of two columns of each of the at least one pair of columns of the memory cell array; and switching output data of the two columns in each of the at least one pair of columns of the memory cell array so as to determine the values on the first neurons of the first layer.

Patent Agency Ranking