A computer-implemented method and computer processing system are provided for update management for a neural network. The method includes performing an isotropic update process on the neural network using a Resistive Processing Unit. The isotropic update process uses a multiplicand and a multiplier from a multiplication operation. The performing step includes scaling the multiplicand and the multiplier to have a same order of magnitude.

A multiplication device for performing a matrix-vector-multiplication may be provided. The multiplication device comprises a memristive crossbar array comprising a plurality of memristive devices. The device comprises a decomposition unit adapted for decomposing a matrix into a partial sum of multiple sub-matrices, and decomposing a vector into a sum of multiple sub-vectors, a programming unit adapted for programming the plurality of the memristive devices with values representing elements of the sub-matrices such that each one of the memristive devices corresponds to one of the elements of the sub-matrices, an applying unit adapted for applying elements of one of the multiple sub-vectors as input values to the memristive crossbar array to input lines of the memristive crossbar array resulting in partial results at output lines of the memristive crossbar array, and a summing unit adapted for scaling and summing the partial results building the product of the matrix and the vector.

A circuit for a neuron of a multi-stage compute process is disclosed. The circuit comprises a weighted charge packet (WCP) generator. The circuit may also include a voltage divider controlled by a programmable resistance component (e.g., a memristor). The WCP generator may also include a current mirror controlled via the voltage divider and arrival of an input spike signal to the neuron. WCPs may be created to represent the multiply function of a multiply accumulate processor. The WCPs may be supplied to a capacitor to accumulate and represent the accumulate function. The value of the WCP may be controlled by the length of the spike in signal times the current supplied through the current mirror. Spikes may be asynchronous. Memristive components may be electrically isolated from input spike signals so their programmed conductance is not affected. Positive and negative spikes and WCPs for accumulation may be supported.

A neural network circuit that uses a ramp function as an activation function includes a memory device in which memristors serving as memory elements are connected in a matrix. The neural network circuit further includes I-V conversion amplification circuits for converting currents flowing via the memory elements into voltages, a differential amplifier circuit for performing a differential operation on outputs of two I-V conversion amplification circuits, an A-D converter for performing an A-D conversion on a result of the differential operation, and an output determine that, by referring to input signals of the differential amplifier circuit, determines whether an output signal value of the differential amplifier circuit belongs to an active region or an inactive region. Based on a determination result, the input determiner switches over the differential amplifier circuit and the A-D converter between an operating state and a standby state.

A circuit includes an engine to compute analog multiplication results between vectors of a sub-matrix. An analog to digital converter (ADC) generates a digital value for the analog multiplication results computed by the engine. A shifter shifts the digital value of analog multiplication results a predetermined number of bits to generate a shifted result. An adder adds the shifted result to the digital value of a second multiplication result to generate a combined multiplication result.

An in-memory computing system for computing vector-matrix multiplications includes an array of resistive memory devices arranged in columns and rows, such that resistive memory devices in each row of the array are interconnected by a respective wordline and resistive memory devices in each column of the array are interconnected by a respective bitline. The in-memory computing system also includes an interface circuit electrically coupled to each bitline of the array of resistive memory devices and computes the vector-matrix multiplication between an input vector applied to a given set of wordlines and data values stored in the array. For each bitline, the interface circuit receives an output in response to the input being applied to the given wordline, compares the output to a threshold, and increments a count maintained for each bitline when the output exceeds the threshold. The count for a given bitline represents a dot-product.

A circuit of cells used as a memory array and capable of in-memory arithmetic which includes a plurality of signed ternary processing, each signed ternary processing cell includes a first memory cell, adapted to hold a first digital value, a second memory cell, adapted to hold a second digital value, wherein a binary combination of the first digital value and the second digital value establishes a first signed ternary operand, a signed ternary input forming a second signed ternary operand, and a signed ternary output, wherein the signed ternary output represents a signed multiplication of the first signed ternary operand and the second signed ternary operand, a sense circuit adapted to output a subtraction result.

A method, computer program product, and circuit are provided for noise and bound management for a Resistive Processing Unit (RPU) array having an op-amp. The method includes reducing the noise in an output signal from the RPU array by using a largest value, in a sigma vector having a plurality of values, as a representation for a window for an input signal to the RPU array. The input signal to the RPU array is formed from the plurality of values. The method further includes sensing saturation at an output of the op-amp. The method also includes managing the bound to eliminate the saturation by reducing the plurality of values from which the input signal to the RPU is formed.

According to an example, discrete-time analog filtering may include receiving an input signal, and sampling the input signal to determine sampled input signal values related to the input signal.

A method, computer program product, and circuit are provided for noise and bound management for a Resistive Processing Unit (RN) array having an op-amp. The method includes reducing the noise in an output signal from the RPU array by using a largest value, in a sigma vector having a plurality of values, as a representation for a window for an input signal to the RPU array. The input signal to the RPU array is formed from the plurality of values. The method further includes sensing saturation at an output of the op-amp. The method also includes managing the bound to eliminate the saturation by reducing the plurality of values from which the input sign to the RPU is formed.