Lossy tensor compression and decompression circuits compress and decompress tensor elements based on the values of neighboring tensor elements. The lossy compression circuit scales each decompressed tensor element of a tile by a scaling factor that is based on the maximum value that can be represented by the number of bits used to represent a compressed tensor element, and the greatest value and least value of the tensor elements of the tile. The lossy decompression circuit performs the inverse of the lossy compression. The compression circuit and decompression circuit have parallel multiplier circuits and parallel adder circuits to perform the lossy compression and lossy decompression, respectively.

Lossy tensor compression and decompression circuits compress and decompress tensor elements based on the values of neighboring tensor elements. The lossy compression circuit scales each decompressed tensor element of a tile by a scaling factor that is based on the maximum value that can be represented by the number of bits used to represent a compressed tensor element, and the greatest value and least value of the tensor elements of the tile. The lossy decompression circuit performs the inverse of the lossy compression. The compression circuit and decompression circuit have parallel multiplier circuits and parallel adder circuits to perform the lossy compression and lossy decompression, respectively.

An electronic device including a neural network processor and a presorter is described. The presorter determines a sorted order to be used by the neural network processor for processing a set of instances of input data through the neural network, the determining including rearranging an initial order of some or all of the instances of input data so that instances of input data having specified similarities among the some or all of the instances of input data are located nearer to one another in the sorted order. The presorter provides, to the neural network processor, the sorted order to be used for controlling an order in which instances of input data from among the set of instances of input data are processed through the neural network. A controller in the electronic device adjusts operation of the presorter based on efficiencies of the presorter and the neural network processor.

An electronic device including a neural network processor and a presorter is described. The presorter determines a sorted order to be used by the neural network processor for processing a set of instances of input data through the neural network, the determining including rearranging an initial order of some or all of the instances of input data so that instances of input data having specified similarities among the some or all of the instances of input data are located nearer to one another in the sorted order. The presorter provides, to the neural network processor, the sorted order to be used for controlling an order in which instances of input data from among the set of instances of input data are processed through the neural network. A controller in the electronic device adjusts operation of the presorter based on efficiencies of the presorter and the neural network processor.

According to one embodiment, an inference apparatus includes a processor. The processor generates an intermediate signal by processing an input signal with a convolutional neural network. The processor extracts one or more intermediate partial signals each serving as part of the intermediate signal from the intermediate signal. The processor calculates a statistic of the one or more intermediate partial signals. The processor outputs an inference result relating to the input signal and corresponding to the statistic.

According to one embodiment, an inference apparatus includes a processor. The processor generates an intermediate signal by processing an input signal with a convolutional neural network. The processor extracts one or more intermediate partial signals each serving as part of the intermediate signal from the intermediate signal. The processor calculates a statistic of the one or more intermediate partial signals. The processor outputs an inference result relating to the input signal and corresponding to the statistic.

A neural network device includes a floating-point arithmetic circuit configured to perform a dot product operation and an accumulation operation; and a buffer configured to store first cumulative data generated by the floating-point arithmetic circuit, wherein the floating-point arithmetic circuit is further configured to perform the dot product operation and the accumulation operation by: identifying a maximum value from a plurality of exponent addition results, obtained by respectively adding exponents of a plurality of floating-point data pairs, and an exponent value of the first cumulative data; performing, based on the maximum value, an align shift of a plurality of fraction multiplication results, obtained by respectively multiplying fractions of the plurality of floating-point data pairs, and a fraction part of the first cumulative data; and performing a summation of the plurality of aligned fraction multiplication results and the aligned fraction part of the first cumulative data.

A neural network device includes a floating-point arithmetic circuit configured to perform a dot product operation and an accumulation operation; and a buffer configured to store first cumulative data generated by the floating-point arithmetic circuit, wherein the floating-point arithmetic circuit is further configured to perform the dot product operation and the accumulation operation by: identifying a maximum value from a plurality of exponent addition results, obtained by respectively adding exponents of a plurality of floating-point data pairs, and an exponent value of the first cumulative data; performing, based on the maximum value, an align shift of a plurality of fraction multiplication results, obtained by respectively multiplying fractions of the plurality of floating-point data pairs, and a fraction part of the first cumulative data; and performing a summation of the plurality of aligned fraction multiplication results and the aligned fraction part of the first cumulative data.

A semiconductor device includes a selection signal generation circuit configured to generate a selection signal by comparing a first input signal and a second input signal. The semiconductor device also includes a comparison signal generation circuit configured to output a comparison signal by selecting one of the first input signal and the second input signal based on the selection signal.

A filter multiplexer for variable bandwidth annealing selection is described. The filter multiplexer has multiple pathways, where each pathway comprises a switch and a filter. Each filter has a different cutoff frequency from the other filters. Switches may be cryogenic switches. Each pathway may be communicatively coupled to an external annealing line. Upon receiving a problem, an annealing bandwidth can be selected, set or configured via the multiplexer to operate a quantum processor with a desired annealing schedule. The multiplexer may be used for calibration of a quantum processor by performing a calibration with a large annealing bandwidth, then calibrating the quantum processor by iterating through all available annealing bandwidths from the multiplexer.