Abstract: An apparatus and method are provided for performing bit permutation operations. The apparatus has an interface for receiving an input data operand and a control operand. The input data operand comprises one or more data elements, each data element comprising a plurality of bits, and the control operand provides control information identifying bit permutations required when performing a given bit permutation operation on each data element. The bit permute circuitry treats the input data operand as a plurality of fixed size data portions, each data element comprising one or more of the data portions with the number being dependent on the data element size. The bit permute circuitry performs bit permutation operations on each data portion of the input data operand, using the control information provided for that data portion, generating, for each data portion, at least one intermediate result.

Abstract: A resampling method based on window function for flexible sampling rate conversion in broadband frequency measurement devices is described. The resampling algorithm can satisfy the requirements of different sampling rates. The frequency responses of the filter in the resampling model based on the Farrow structure are analyzed, and the design criterion of the filter in resampling model is considered. A fractional delay filter design model based on window function method is described. A fractional delay filter matrix, which is expressed by polynomial form, is constructed. Then the expression related to subfilter coefficients is obtained and subfilter coefficients are solved for by the least square method.

Abstract: The base-2n in-memory adder device mainly comprises Perpetual Digital Perceptron (PDP) in-memory adder with Read Only Memory (ROM) arrays for storing the binary sum codes of the addition table for processing the addition operations of two n-bit binary integer operands. Since the integer numbers can be represented by the binary codes of multiple digits of base-2n integer numbers, the base-2n in-memory adder device can iterate multiple times of the digit-additions to complete the binary code addition for two m-digit base-2n integer operands. Consequently, the base-2n in-memory adder device can improve the computation efficiency and save the computation power by eliminating the data transportations between Arithmetic Logic Unit (ALU), registers, and memory units.

Abstract: Top-k is a process by which the largest elements among a set of elements is found. In various implementations, a top-k computation can be executed by a neural network accelerator, where the top-k computation is performed using a process that makes use of the accelerators memory array. A set of numerical values on which to perform top-k can be stored in the memory array. The accelerator can locate the maximum value from among the set of numerical values, and can store the maximum value back into the memory array. The accelerator can next remove the maximum value from the set of numerical values, so that a next largest value can be found. To remove the maximum value, the accelerator can write a value representing negative infinity to the memory array at each location of the maximum value.

Abstract: Validating microprocessor instruction execution by receiving a floating-point exception selection, receiving a validation method selection, generating validation data according to the floating-point exception selection and the validation method selection by randomly generating a first tensor element value and randomly generating a second tensor element value according to the first tensor element value and the floating-point exception selection, and executing a floating-point computation according to the validation data.

Abstract: Multiple random numbers are generated. The multiple random numbers are N different random numbers. N is a positive integer. Generating the multiple random numbers includes generating a random number array including N storage units. The multiple random numbers are shuffled. A random number obtaining instruction is received. A random number is obtained from the multiple random numbers based on the random number obtaining instruction.

Abstract: A graphics processing unit (GPU) schedules recurrent matrix multiplication operations at different subsets of CUs of the GPU. The GPU includes a scheduler that receives sets of recurrent matrix multiplication operations, such as multiplication operations associated with a recurrent neural network (RNN). The multiple operations associated with, for example, an RNN layer are fused into a single kernel, which is scheduled by the scheduler such that one work group is assigned per compute unit, thus assigning different ones of the recurrent matrix multiplication operations to different subsets of the CUs of the GPU. In addition, via software synchronization of the different workgroups, the GPU pipelines the assigned matrix multiplication operations so that each subset of CUs provides corresponding multiplication results to a different subset, and so that each subset of CUs executes at least a portion of the multiplication operations concurrently.

Abstract: A filter circuit includes a first stage comprising a first infinite impulse response (IIR) filter; a third stage comprising a third IIR filter; and a second stage interposed between the first stage and the third stage, the second stage comprising a second IIR filter, where an output terminal of the first IIR filter is coupled to an input terminal of the second IIR filter, and an output terminal of the second IIR filter is coupled to an input terminal of the third IIR filter, where the second stage of the filter circuit is configured to operate in an acquisition mode when a transient is detected in an input signal to the first IIR filter, where during the acquisition mode, the second stage of the filter circuit is bypassed.

Abstract: Systems, apparatuses, methods, and computer program products are disclosed for quantum random number generation (QRNG). An example method includes generating, by encoding circuitry of a QRNG chip, a set of time-bin qubits based on a first optical path length. The example method further includes transmitting, by the encoding circuitry of the QRNG chip, the set of time-bin qubits over an optical line. The example method further includes receiving, by decoding circuitry of the QRNG chip, the set of time-bin qubits over the optical line. The example method further includes measuring, by the decoding circuitry of the QRNG chip and based on a second optical path length different from the first optical path length, the set of time-bin qubits to generate a decoded set of bits. In some embodiments, the example method further includes generating, by session authentication circuitry, a session key based on the decoded set of bits.

Abstract: Herein are disclosed computation units for element approximation. A computation unit may include a first circuit to compute a first projection ? of an input element xi from a first range to a second range. In the first circuit, the input element xi may have a first format and the projected element yi may have a second format. In addition, in the first circuit, the second format may have more bits than the first format. The computation unit may further include a second circuit operatively coupled to the first circuit to produce a reduction zi in the first format using the projected element yi in the second format. The computation unit may also include a third circuit operatively coupled to the second circuit to compute a second projection ? of the reduction zi from the second range to the first range to produce an approximation wi.

Abstract: A system includes an accelerator to accelerate the computations of nonlinear math functions. The accelerator includes a set of first evaluators and a set of second evaluators. Each of the first evaluators and the second evaluators is a fixed-function hardware circuit specialized for evaluating a respective nonlinear function. The system further includes a processor which decodes a math instruction in an instruction set, sends to the accelerator a signal identifying a nonlinear function corresponding to the math instruction, and receives an output of the nonlinear function from the accelerator. According to the signal, the accelerator evaluates the nonlinear function using one of the first evaluators. When the signal identifies the nonlinear function as a composite function, the accelerator additionally uses one of second evaluators on an output of the one first evaluator.

Abstract: A computer program product, an apparatus, a functionally safe programmable controller and a method for processing data, wherein an uncoded real number x is converted into a logarithmic number system (LNS) coded integer xLNS via a predetermined conversion rule for a logarithmic number system (LNS) in accordance with the relationship: xLNS=sgn(x)·2m+Id|x|·2n, where sgn(x) denotes a sign function of the uncoded real number x, Id|x| denotes a binary logarithm of the uncoded real number x, m denotes a first exponent and n denotes a second exponent, and the LNS-coded integer xLNS is coded into an arithmetically coded integer xc via arithmetic coding such that the required integer operations is reduced.

Abstract: A receiver circuit is disclosed. The receiver circuit includes a multi-PAM input circuit to receive a multi-PAM input symbol. The input symbol exhibits one of multiple threshold levels during a sampling period. The threshold levels correspond to a set of M-bit two's-complement values within a defined set of threshold values. An adaptive filtering circuit includes a first transcoder to transcode the set of M-bit two's-complement values to a set of N-bit values, where N<M. An adaptive filter operates to filter the set of N-bit values to generate a filtered set of data values. A second transcoder transforms the filtered set of data values to a second set of data values that corresponds to a set of filtered M-bit two's-complement values.

Abstract: Configurable specialized processing blocks, such as DSP blocks, are described that implement fixed and floating-point functionality in a single mixed architecture on a programmable device. The described architecture reduces the need to construct floating-point functions outside the configurable specialized processing block, thereby minimizing hardware cost and area. The disclosed architecture also introduces pipelining into the DSP block in order to ensure the floating-point multiplication and addition functions remain in synchronicity, thereby increasing the maximum frequency at which the DSP block can operate. Moreover, the disclosed architecture includes logic circuitry to support floating-point exception handling.

Abstract: Techniques are described herein for performing efficient matrix multiplication in architectures with scratchpad memories or associative caches using asymmetric allocation of space for the different matrices. The system receives a left matrix and a right matrix. In an embodiment, the system allocates, in a scratchpad memory, asymmetric memory space for tiles for each of the two matrices as well as a dot product matrix. The system proceeds with then performing dot product matrix multiplication involving the tiles of the left and the right matrices, storing resulting dot product values in corresponding allocated dot product matrix tiles. The system then proceeds to write the stored dot product values from the scratchpad memory into main memory.

Abstract: This disclosure describes techniques for analyzing statistical quality of bitstrings produced by a physical unclonable function (PUF). The PUF leverages resistance variations in the power grid wires of an integrated circuit. Temperature and voltage stability of the bitstrings are analyzed. The disclosure also describes converting a voltage drop into a digital code, wherein the conversion is resilient to simple and differential side-channel attacks.

Abstract: A data processing apparatus is provided to convert a plurality of signed digits to an output value. Receiver circuitry receives, at each of a plurality of iterations, one of the plurality of signed digits, each of the signed digits comprising a number of bits dependent on a radix. The signed digits being used to form an unrounded output value followed by zero or more extra bits. Adjustment circuitry adjusts a least-significant digit of the unrounded output value to produce an incremented unrounded output value after the plurality of iterations. Rounding circuitry selects from among the unrounded output value and the incremented unrounded output value to produce the output value. The adjustment circuitry is adapted, when a value of a position of a least-significant bit of the unrounded output value is greater than or equal to the radix divided by two, to adjust a subset of the digits of the unrounded output value.

Abstract: A multi-addend adder circuit used for multi-addend addition in a polar representation in stochastic computing. The multi-addend adder circuit includes a buffer circuit and a computing circuit, where the buffer circuit is configured to store to-be-buffered data for at least one cycle and output buffer data, and the computing circuit is configured to process a plurality of pieces of bitstream data and the buffer data and output one piece of bitstream data and the to-be-buffered data, where the piece of output bitstream data is a quotient of dividing a sum of summation data and the buffer data by a scale-down coefficient, the output to-be-buffered data is a remainder of dividing a sum of all summation data until a current cycle by the scale-down coefficient, and the summation data is a quantity of bits whose values are 1 in the plurality of pieces of first bitstream data.

Abstract: An apparatus and method are provided for performing an index operation. The apparatus has vector processing circuitry to perform an index operation in each of a plurality of lanes of parallel processing. The index operation requires an index value opm to be multiplied by a multiplier value e to produce a multiplication result. The number of lanes of parallel processing is dependent on a specified element size, and the multiplier value is different, but known, for each lane of parallel processing. The vector processing circuitry comprises mapping circuitry to perform, within each lane, mapping operations on the index value opm in order to generate a plurality of intermediate input values. The plurality of intermediate input values are such that the addition of the plurality of intermediate input values produces the multiplication result. Within each lane the mapping operations are determined by the multiplier value used for that lane.

Abstract: An electronic apparatus is provided. The electronic apparatus includes a storage storing object data and kernel data, and a processor including a plurality of processing elements arranged in a matrix formation, wherein the processor is configured to input corresponding first elements among a plurality of first elements included in the object data into processing elements arranged in a first row among the plurality of processing elements, and input a plurality of second elements included in the kernel data sequentially into the processing elements arranged in the first row to perform operations between the corresponding first elements and the plurality of second elements, to identify a depth in which a first element and a second element have a non-zero value, and to input the first element and the second element corresponding to the identified depth into a calculator included in each of the processing elements arranged in the first row to perform a convolution operation.