Abstract: Embodiments herein relate to a neural network processor in a control loop, where the control loop sets an optimum supply voltage for the processor based on a measured error count or rate of the neural network. For example, if the measured error count is greater than a target level or range, the supply voltage can be increased. If the measured error count is below the target level or range, the supply voltage can be decreased. The error rate can be measured by providing an error detection circuit for one or more monitored nodes/processing units of a neural network. The error detection circuit can receive the same input data as the associated monitored processing unit, but operates on only a portion of the input data.

Abstract: Systems, apparatuses and methods include technology that receives, with a first plurality of multipliers of a multiply-accumulator (MAC), first digital signals from a memory array, wherein the first plurality of multipliers includes a plurality of capacitors. The technology further executes, with the first plurality of multipliers, multibit computation operations with the plurality of capacitors based on the first digital signals, and generates, with the first plurality of multipliers, a first analog signal based on the multibit computation operations.

Abstract: Systems, apparatuses and methods include technology that receives, with a first plurality of multipliers of a multiply-accumulator (MAC), first digital signals from a memory array, wherein the first plurality of multipliers includes a plurality of capacitors. The technology further executes, with the first plurality of multipliers, multibit computation operations with the plurality of capacitors based on the first digital signals, and generates, with the first plurality of multipliers, a first analog signal based on the multibit computation operations.

Abstract: Technology for generating an SRAM-based in-memory computing macro includes replacing a SRAM cell cluster defined by a generic SRAM macro with a single-bit multi-bank cluster, the single-bit multi-bank cluster including a plurality of CiM SRAM cells and a plurality of C-2C capacitor ladder cells, arranging a plurality of single-bit multi-bank clusters to form a multi-bit multi-bank cluster, and arranging a plurality of multi-bit multi-bank clusters into a multi-dimensional MAC computational unit within a region of the generic SRAM macro, where an output of at least two of the multi-bit multi-bank clusters are electrically coupled to form an output analog activation line, and where a plurality of bit lines and a plurality of word lines remain at the same grid locations as provided in the generic SRAM macro. Embodiments include arranging a plurality of multi-dimensional MAC computational units into an in-memory MAC computing array.

Abstract: Technology for generating an SRAM-based in-memory computing macro includes replacing a SRAM cell cluster defined by a generic SRAM macro with a single-bit multi-bank cluster, the single-bit multi-bank cluster including a plurality of CiM SRAM cells and a plurality of C-2C capacitor ladder cells, arranging a plurality of single-bit multi-bank clusters to form a multi-bit multi-bank cluster, and arranging a plurality of multi-bit multi-bank clusters into a multi-dimensional MAC computational unit within a region of the generic SRAM macro, where an output of at least two of the multi-bit multi-bank clusters are electrically coupled to form an output analog activation line, and where a plurality of bit lines and a plurality of word lines remain at the same grid locations as provided in the generic SRAM macro. Embodiments include arranging a plurality of multi-dimensional MAC computational units into an in-memory MAC computing array.

Abstract: An analog to digital converter (ADC) circuit including: a voltage-controlled delay circuit (VCDC) configured to: sample a received analog signal based on an input clock signal to generate analog signal samples; generate an output signal representative of the input clock signal shifted in the time domain with a delay based on the analog signal samples; and a time to digital converter (TDC) coupled to the voltage-controlled delay circuit and configured to generate a digital output signal based on the output signal.

Abstract: An example optical neural network includes a first layer having a laser responsive to an input signal to transmit an optical signal, a second layer having a photodetector to generate an electrical signal based on the optical signal, and a third layer having a memory array to store weights of the optical neural network, the third layer to generate an output signal based on the electrical signal and at least one of the weights.

Abstract: Some challenges to using analog compute-in-memory circuits for machine learning hardware relate to the overhead and non-idealities associated with data converters at the input and output of the analog compute-in-memory circuits. To address at least some of these challenges, a digital-to-analog converter having binary-weighted resistances can be used to drive the analog compute-in-memory circuits. The resulting digital-to-analog converter is sparsity-aware with low average power consumption. A calibration engine can perform analog tuning and/or digital post-correction to mitigate the non-idealities of the digital-to-analog converter.

Abstract: Techniques are disclosed for implementing a CMOS-compatible millimeter wave matrix computing network architecture, which enables high-speed matrix operations for deep learning neural networks through a reconfigurable feedforward architecture using matrix computing meshes. Each mesh may include hybrid couplers and adjustable phase shifters. The architecture may be configured in various arrangements with programmable weights. The architecture offers advantages over existing solutions through full CMOS compatibility, the elimination of optical-electrical conversion, improved scalability, total latency, and superior power efficiency. Applications include massive MIMO systems and cognitive radar, in which the network may be implemented as part of RF front ends to reduce ADC requirements, system complexity, and power consumption.

Abstract: A radio communication system includes an antenna array comprising a plurality of antenna elements configured to receive a plurality of radio frequency (RF) signals, RF circuitry coupled to the antenna array configured to downconvert the RF signals to generate analog baseband signals, and analog processing circuitry coupled to the RF circuitry. The analog processing circuitry is configured to generate spatially compressed beamspace domain analog signals from the analog baseband signals.

Abstract: A wireless communication device, including a radiofrequency frontend, configured to wirelessly receive a radiofrequency signal; perform one or more analog baseband operations on the received radiofrequency signal, according to a radio access technology; and output an analog signal representing an output of the analog baseband operations on the received radiofrequency signal; an error corrector, configured to perform an error correction operation on the analog signal; and output an error corrected signal in analog domain; and the analog-digital converter, configured to convert the error corrected signal to digital domain.

Abstract: Systems, methods, apparatuses, and computer-readable media. An analog router of a first supertile of a plurality of supertiles of a network on a chip (NoC) may receive a first analog output from a first compute-in-memory tile of a plurality of compute-in-memory tiles of the first supertile. The analog router may determine, based on a configuration of a neural network executing on the NoC, that a destination of the first analog output includes a second supertile of the plurality of supertiles. An analog-to-digital converter (ADC) of the analog router may convert the first analog output to a first digital output and transmit the first digital output to the second supertile via a communications bus of the NoC.

Abstract: Systems, apparatuses and methods may provide for technology including a digital to analog conversion (DAC) stage to generate analog input activation signals, a multiply-accumulate (MAC) computation stage coupled to the DAC stage, the MAC computation stage to generate output activation results based on the analog input activation signals and multi-bit weight data stored in the MAC computation stage, an analog integration stage coupled to the MAC computation stage, the analog integration stage to conduct partial sum accumulations on the output activation results, and analog to digital conversion (ADC) stage coupled to the analog integration stage, the ADC stage to generate digital computation results based on an output of the analog integration stage, and a controller to vary a number of cycles in the partial sum accumulations based on an overflow condition associated with one or more of the output activation results or the output of the analog integration stage.

Abstract: Various embodiments provide apparatuses, systems, and methods for multibit analog representation, e.g., for in-memory computing. Embodiments may include a single-ended or differential ladder network to generate an analog value (e.g., a voltage or charge) based on a set of bits from a memory array. The ladder network may include a plurality of branches coupled to an output line, wherein individual branches include a capacitor with a first terminal coupled to the output line and a switch coupled to a second terminal of the capacitor. The switch may be controlled by a respective bit of the set of bits to selectively couple the second terminal of the capacitor to a first voltage node or a second voltage node based on a value of the respective bit. Other embodiments may be described and claimed.

Abstract: Methods, apparatus, systems, and articles of manufacture providing an improved Bayesian neural network and methods and apparatus to operate the same are disclosed. An example apparatus includes an oscillator to generate a first clock signal; a resistive element to adjust a slope of a rising edge of a second clock signal; a voltage sampler to generate a sample based on at least one of (a) a first voltage of the first clock signal when a second voltage of the second clock signal satisfies a threshold or (b) a third voltage of the second clock signal when a fourth voltage of the first clock signal satisfies the threshold; and a charge pump to adjust a weight based on the sample, the weight to adjust data in a model.

Abstract: Systems, apparatuses and methods may provide for technology that includes a capacitor ladder, a plurality of memory cells coupled to the capacitor ladder, the plurality of memory cells to control the capacitor ladder to conduct multi-bit multiply accumulate (MAC) operations during a computation phase, and a successive approximation register (SAR) coupled to the capacitor ladder, the SAR to control the capacitor ladder to digitize results of the multi-bit MAC operations during a digitization phase.

Abstract: A radiofrequency frontend device includes a memory array, which includes a plurality of input lines; a plurality of output lines; and a plurality of impedance devices, each impedance device connecting an input line of the plurality of input lines to an output line of the plurality of output lines, wherein each impedance represents a filter coefficient; wherein the radiofrequency frontend device is configured to provide at each input line of the plurality of input lines a sampled voltage of an analog electric signal, each sampled voltage corresponding to a voltage of the analog electric signal during a respective time period of a plurality of time periods; and when the memory array receives the sampled voltages, the memory array is configured to modify each of the sampled voltages by a respective impedance device of the plurality of impedance devices and sum the modified sampled voltages.

Abstract: Circuitry for a compute-in-memory (CiM) circuit or structure arranged to detect bit errors in a group of memory cells based on a summation of binary 1's included in at least one weight matrix stored to the group of memory cells, a parity value stored to another group of memory cells and a comparison of the summation or the parity value to an expected value.

Abstract: Systems, apparatuses and methods include technology that identifies workload numbers associated with a workload. The technology converts the workload numbers to block floating point numbers based on a division of mantissas of the workload numbers into sub-words and executes a compute-in memory operation based on the sub-words to generate partial products.

Abstract: A compute-in-memory (CiM) circuit or structure arranged to detect errors. Examples include detecting errors associated with weight bits stored to computational nodes included in a CiM circuit or structure based on use of complimented bit values. Examples also include detecting errors in the CiM circuit or structure based on using at least some computational nodes included in an array of computational nodes to monitor for the errors during generation of computation results by other computational nodes included in the array.