Abstract: An arithmetic device includes a multiplying-accumulating (MAC) operator and an activation function (AF) circuit. The MAC operator performs a MAC arithmetic operation for weight data and vector data to generate an arithmetic result signal. The AF circuit extracts a first bit group and a second bit group from the arithmetic result signal. In addition, the AF circuit generates an input distribution signal based on the first bit group and the second bit group. Moreover, the AF circuit selects and outputs an output distribution signal that corresponds to the input distribution signal based on an activation function.

Abstract: A random number generator circuit includes a noise source capable of providing a noise signal that varies randomly; and a circuit for extracting the noise signal including an edge detector configured to produce from the noise signal an analogue signal including voltage pulses, each voltage pulse corresponding to a rising or falling edge of the noise signal, and an analogue-to-digital converter configured to generate a random bit sequence from the analogue signal.

Abstract: A system, method, and computer program product are disclosed. The method includes loading a first set of data as an initial matrix and determining a truncated singular value decomposition (SVD) of the initial matrix. The method also includes loading a second set of data as a new matrix, generating a first projection matrix, which approximates k leading left singular vectors of the updated matrix, and generating a second projection matrix, which approximates k leading right singular vectors of the updated matrix. Further, the method includes determining based on the initial matrix, the new matrix, the SVD of the existing matrix, and the first or second projection matrix, an approximate truncated SVD of the updated matrix.

Abstract: Techniques are disclosed relating to dedicated power function circuitry for a floating-point power instruction. In some embodiments, execution circuitry is configured to execute a floating-point power instruction to evaluate the power function xy as 2y log2x. In some embodiments, base-2 logarithm circuitry is configured to evaluate a base-2 logarithm for a first input (e.g., log2 x) by determining coefficients for a polynomial function and evaluating the polynomial function using the determined coefficients and the first input. In some embodiments, multiplication circuitry multiplies the base-2 logarithm result by a second input to generate a multiplication result. In some embodiments, base-2 power function circuitry is configured to evaluate a base-2 power function for the multiplication result. Disclosed techniques may advantageously increase performance and reduce power consumption of floating-point power function operations with reasonable area and accuracy, relative to traditional techniques.

Abstract: A processing device having a sequence of sorting elements arranged in an array. Each of the sorting elements stores a previously retained value therein and receives an input value from a previous sorting element. Each sorting element applies retention logic to select one of the input value or the retained value to be passed to the next sorting element in the array. The value that is passed to the next sorting element can either be set to be the larger, or the smaller, of the input value and the previously retained value, as desired. Rows of processing elements in the array operate in parallel such that large data streams are sorted in parallel (with the data values moving down from one row of processing elements to the next row such that the largest, or the smallest, data values accumulating in the final row of processing elements).

Abstract: A calculation apparatus according to an embodiment includes matrix multiplication circuitry, time evolution circuitry, management circuitry, and output circuitry. The matrix multiplication circuitry calculates N second intermediate variables at a first time point by matrix multiplication between N (N>=2) first intermediate variables at the first time point and a preset coefficient matrix in N rows and N columns. The time evolution circuitry calculates N first variables at a second time point and N first intermediate variables at the second time point, the second time point being a time point following one sampling period after the first time point. The management circuitry increments time point from a start time point for each sampling period and controls the above circuitry to perform a process for each time point. The output circuitry outputs N first variables at a preset end time point.

Abstract: A dynamic bias analog vector-matrix multiplication operation circuit comprises: positive value weight columns (101-10N), constant columns (201-20M) and subtractors (301-30N), wherein the number of the subtractors is equal to the number of the positive value weight columns, the subtractors are correspondingly connected to the positive value weight columns on a one-to-one basis, and the number of the constant columns is less than the number of the positive value weight columns; minuend input ends of the subtractors are correspondingly connected to output ends of the positive value weight columns, subtrahend input ends of a plurality of subtractors are connected to the same constant column, and output ends thereof output operation results. Before a weight is written in a programmable semiconductor device, a constant positive value is added to each element in a weight array, the weight array is written in a positive value weight column, and the constant positive value is written in a constant column.

Abstract: An actively stabilized random number generator includes a random number generator and a feedback controller. The random number generator includes a chaotic physical circuit realizing an iterated function. The iterated function is configured to produce a trajectory of iterates and has an operating parameter ? and a desired Markov operating point. A binary bit converter has a symbol function configured to produce binary symbols from the trajectory of iterates and a maximal kneading sequence. The feedback controller is configured to observe the maximal kneading sequence within the trajectory of iterates and adjust the operating parameter to the desired Markov operating point.

Abstract: A batched Cholesky decomposition method, system, and non-transitory computer readable medium for a Graphics Processing Unit (GPU), include mirroring matrices to form paired matrices solving the paired matrices simultaneously.

Abstract: An integrated circuit with specialized processing blocks are provided. A specialized processing block may be optimized for machine learning algorithms and may include a multiplier data path that feeds an adder data path. The multiplier data path may be decomposed into multiple partial product generators, multiple compressors, and multiple carry-propagate adders of a first precision. Results from the carry-propagate adders may be added using a floating-point adder of the first precision. Results from the floating-point adder may be optionally cast to a second precision that is higher or more accurate than the first precision. The adder data path may include an adder of the second precision that combines the results from the floating-point adder with zero, with a general-purpose input, or with other dot product terms. Operated in this way, the specialized processing block provides a technical improvement of greatly increasing the functional density for implementing machine learning algorithms.

Abstract: In a system with control logic and a processing element array, two modes of operation may be provided. In the first mode of operation, the control logic may configure the system to perform matrix multiplication or 1×1 convolution. In the second mode of operation, the control logic may configure the system to perform 3×3 convolution. The processing element array may include an array of processing elements. Each of the processing elements may be configured to compute the dot product of two vectors in a single clock cycle, and further may accumulate the dot products that are sequentially computed over time.

Abstract: Detailed are embodiments related to bit matrix multiplication in a processor. For example, in some embodiments a processor comprising: decode circuitry to decode an instruction have fields for an opcode, an identifier of a first source bit matrix, an identifier of a second source bit matrix, an identifier of a destination bit matrix, and an immediate; and execution circuitry to execute the decoded instruction to perform a multiplication of a matrix of S-bit elements of the identified first source bit matrix with S-bit elements of the identified second source bit matrix, wherein the multiplication and accumulation operations are selected by the operation selector and store a result of the matrix multiplication into the identified destination bit matrix, wherein S indicates a plural bit size is described.

Abstract: An optical computation system, preferably including an optical source, a splitter, and one or more phase accumulator banks. A phase accumulator bank, preferably including two optical paths, a plurality of phase accumulator units, and a detector module, and optionally including one or more compensation phase shifters. A method, preferably including receiving one or more optical inputs, receiving one or more electrical inputs, controlling one or more phase accumulator units based on the electrical inputs, and generating one or more electrical outputs based on optical signals.

Abstract: A method is provided that includes performing, by a processor in response to a floating point multiply instruction, multiplication of floating point numbers, wherein determination of values of implied bits of leading bit encoded mantissas of the floating point numbers is performed in parallel with multiplication of the encoded mantissas, and storing, by the processor, a result of the floating point multiply instruction in a storage location indicated by the floating point multiply instruction.

Abstract: Today neural networks are used to enable autonomous vehicles and improve the quality of speech recognition, real-time language translation, and online search optimizations. However, operation of the neural networks for these applications consumes energy. Quantization of parameters used by the neural networks reduces the amount of memory needed to store the parameters while also reducing the power consumed during operation of the neural network. Matrix operations performed by the neural networks require many multiplication calculations, so reducing the number of bits that are multiplied reduces the energy that is consumed. Quantizing smaller sets of the parameters using a shared scale factor improves accuracy compared with quantizing larger sets of the parameters. Accuracy of the calculations may be maintained by quantizing and scaling the parameters using fine-grained per-vector scale factors. A vector includes one or more elements within a single dimension of a multi-dimensional matrix.

Abstract: In a multiply accumulate (MAC) unit, an accumulator may be implemented in two or more stages. For example, a first accumulator may accumulate products from the multiplier of the MAC unit, and a second accumulator may periodically accumulate the running total of the first accumulator. Each time the first accumulator's running total is accumulated by the second accumulator, the first accumulator may be initialized to begin a new accumulation period. In one embodiment, the number of values accumulated by the first accumulator within an accumulation period may be a user-adjustable parameter. In one embodiment, the bit width of the input of the second accumulator may be greater than the bit width of the output of the first accumulator. In another embodiment, an adder may be shared between the first and second accumulators, and a multiplexor may switch the accumulation operations between the first and second accumulators.

Abstract: Processing cores with data associative adaptive rounding and associated methods are disclosed herein. One disclosed processing core comprises an arithmetic logic unit cluster configured to generate a value for a unit of directed graph data using input directed graph data, a comparator coupled to a threshold register and a data register, a core controller configured to load a threshold value into the threshold register when the value for the unit of directed graph data is loaded into the data register, and a rounding circuit. The rounding circuit is configured to receive the value for the unit of directed graph data from the arithmetic logic unit cluster and conditionally round the value for the unit of directed graph data based on a comparator output from the comparator.

Abstract: A data processing circuit is disclosed. The data processing circuit relates to the field of digital circuits, and includes a first computing circuit and an input control circuit. The first computing circuit includes one or more computing sub-circuits. Each computing sub-circuit includes a first addition operation circuit, a multiplication operation circuit, a first comparison operation circuit, and a first nonlinear operation circuit. The first nonlinear operation circuit includes at least one of an exponential operation circuit and a logarithmic operation circuit. The input control circuit is configured to: control the first computing circuit to read input data and an input parameter, and control, according to a received first instruction, the operation circuit in the computing sub-circuit included in the first computing circuit, to perform an operation on the input data and the input parameter.

Abstract: A tile of an FPGA includes a multiple mode arithmetic circuit. The multiple mode arithmetic circuit is configured by control signals to operate in an integer mode, a floating-point mode, or both. In some example embodiments, multiple integer modes (e.g., unsigned, two's complement, and sign-magnitude) are selectable, multiple floating-point modes (e.g., 16-bit mantissa and 8-bit sign, 8-bit mantissa and 6-bit sign, and 6-bit mantissa and 6-bit sign) are supported, or any suitable combination thereof. The tile may also fuse a memory circuit with the arithmetic circuits. Connections directly between multiple instances of the tile are also available, allowing multiple tiles to be treated as larger memories or arithmetic circuits. By using these connections, referred to as cascade inputs and outputs, the input and output bandwidth of the arithmetic circuit is further increased.

Abstract: A system and/or an integrated circuit including: (a) a multiplier-accumulator execution pipeline including multiplier-accumulator circuits to process image data, using associated filter weights, via a plurality of multiply and accumulate operations and (b) first data format conversion circuitry including (i) inputs to receive filter weights of a plurality of sets of filter weights, wherein each set includes a plurality of filter weights each having a block-scaled fraction data format, (ii) conversion circuitry, coupled to the inputs, to convert the filter weights of each set from the block-scaled fraction data format to a floating point data format, and (iii) outputs to output the filter weights having the floating point data format. In operation, the multiplier-accumulator circuits of the pipeline are configured to perform the plurality of multiply and accumulate operations using (a) the image data and (b) the filter weights having the floating point data format.