Memristor-based multipliers using memristors-as-drivers (MAD) gates. As a result of employing MAD gates in memristor-based multipliers, such as shift-and-add multipliers, Booth multipliers and array multipliers, the number of delay steps may be less than half than the number of delay steps required in traditional CMOS implementations of multipliers. Furthermore, by using MAD gates, memristor-based multipliers can be implemented with less complexity (e.g., fewer memristors and drivers). As a result, by the memristor-based multipliers using MAD gates, the speed and complexity of a wide variety of arithmetic operations is improved.

Disclosed techniques provide for dynamically changing precision of a multi-stage compute process. For example, changing neural network (NN) parameters on a per-layer basis depending on properties of incoming data streams and per-layer performance of an NN among other considerations. NNs include multiple layers that may each be calculated with a different degree of accuracy and therefore, compute resource overhead (e.g., memory, processor resources, etc.). NNs are usually trained with 32-bit or 16-bit floating-point numbers. Once trained, an NN may be deployed in production. One approach to reduce compute overhead is to reduce parameter precision of NNs to 16 or 8 for deployment. The conversion to an acceptable lower precision is usually determined manually before deployment and precision levels are fixed while deployed. Disclosed techniques and implementations address automatic rather than manual determination or precision levels for different stages and dynamically adjusting precision for each stage at run-time.

Memristor-based dividers using memristors-as-drivers (MAD) gates. As a result of employing MAD gates in memristor-based dividers, such as binary non-restoring dividers and SRT dividers, the number of delay steps may be less than half than the number of delay steps required in traditional CMOS implementations of dividers. Furthermore, by using MAD gates, memristor-based dividers can be implemented with less complexity (e.g., fewer memristors and drivers). As a result, by the memristor-based dividers using MAD gates, the speed and complexity of a wide variety of arithmetic operations is improved.

Technology is described herein for performing multiplication using non-volatile memory cells. A multiplicand may be stored a node that includes multiple non-volatile memory cells. Each memory cell in a node may be programmed to one of two physical states, with each non-volatile memory cell storing a different bit of the multiplicand. Multiplication may be performed by applying a multiply voltage to the node of memory cells and processing memory cell currents from the memory cells in the node. The memory cell current from each memory cell in the node is multiplied by a different power of two. The multiplied signals are summed to generate a result signal, which represents a product of the multiplier and a multiplicand stored in the node. If desired, binary memory cells may be used to perform multiplication. Vector/vector and vector/matrix multiplication may also be performed.

Examples of the present disclosure provide apparatuses and methods for performing multiplication operations in a memory. An example method comprises performing a multiplication operation on a first element stored in a group of memory cells coupled to a first access line and a number of sense lines of a memory array and a second element stored in a group of memory cells coupled to a second access line and the number of sense lines of the memory array. The method can include a number operations performed without transferring data via an input/output (I/O) line.

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 neural network circuit includes a memory device in which memristors being variable resistance elements are connected in a matrix and serve as memory elements of the memory device. The neural network circuit further includes a voltage application device arranged to apply a bias voltage to the memory device and current-voltage (I-V) conversion amplification circuits arranged to convert currents flowing via the memory elements into voltages and output the voltage. A feedback resistor of a respective I-V conversion amplification circuit includes a memristor. The feedback resistor of a respective I-V conversion amplification circuit and the memory elements acting as an input resistor of the I-V conversion amplification circuit are connected to align a polarity direction of the memristor of the feedback resistor and polarity directions of the memristors of the memory elements acting as the input resistor.

A multiplier device for binary magnetic applied fields uses Interlayer Exchange Coupling (IEC) structure where two layers of ferromagnetic material are separated from each other by non-magnetic layers of electrically conductive material of atomic thickness, sufficient to generate anti-magnetic response in a magnetized layer. A plurality of regions on a top surface are activated with a magnetic field in a first direction for a 1 value and in an opposite direction for a 0 value, the multiplication result presented as magnetic field direction on a plurality of output ferromagnetic regions.

A dot product multiplier for matrix operations for an A matrix of order 1m with a coefficient B matrix of order mm. Processing Elements (PEs) are arranged in an mm array, the columns of the array summed to provide a dot product result. Each of the PEs contains a sign determiner and a plurality of analog multiplier cells, one multiplier cell for each value bit. The multipliers operate over four clock cycles, initializing a capacitor charge according to sign on a first clock phase, sharing charge on a second phase, canceling charge on a third phase, and outputting the resultant charge on a fourth phase, the resultant charge on each column representing the dot product for that column.

According to examples, an apparatus may include an arithmetic logic unit (ALU) to apply a modification function to a digital input signal to generate a modified digital input signal, a digital-to-analog converter (DAC) to convert the modified digital input signal to an analog input signal, a crossbar array to apply an operation on the analog input signal to generate an analog output signal, and an analog-to-digital converter (ADC). The ADC may modify the analog output signal to compensate for application of the modification function to the digital input signal, may convert the modified analog output signal to a digital output signal, and may output the digital output signal.