In described examples, a sample and hold circuit is configured to periodically connect one input of an op-amp to a reference voltage through a switch while a second input of the op-amp is connected to an output of the op-amp. Offset cancellation is performed by storing a sampled offset on a sampling capacitor coupled to the second input of the op-amp.

In described examples, a sample and hold circuit is configured to periodically connect one input of an op-amp to a reference voltage through a switch while a second input of the op-amp is connected to an output of the op-amp. Offset cancellation is performed by storing a sampled offset on a sampling capacitor coupled to the second input of the op-amp.

An apparatus comprises at least one processor and at least one memory including instruction code configured to, with the at least one processor, cause the apparatus at least to perform, with a first accuracy, a first portion of a bitwise multiplication of first and second digital inputs and to perform, with a second accuracy different than the first accuracy, at least a second portion of the bitwise multiplication.

An apparatus comprises at least one processor and at least one memory including instruction code configured to, with the at least one processor, cause the apparatus at least to perform, with a first accuracy, a first portion of a bitwise multiplication of first and second digital inputs and to perform, with a second accuracy different than the first accuracy, at least a second portion of the bitwise multiplication.

Computing-in-memory utilizes memory as weight for multiply-and-accumulate (MAC) operations. Input data multiplies weights to produce output data during the operation. Method of data conversion from input data, memory element to output data is described to enhance the computing efficiency.

Computing-in-memory utilizes memory as weight for multiply-and-accumulate (MAC) operations. Input data multiplies weights to produce output data during the operation. Method of data conversion from input data, memory element to output data is described to enhance the computing efficiency.

Numerous embodiments of analog neural memory arrays are disclosed. In one embodiment, an analog neural memory system comprises an array of non-volatile memory cells, wherein the cells are arranged in rows and columns, the columns arranged in physically adjacent pairs of columns, wherein within each adjacent pair one column in the adjacent pair comprises cells storing W+ values and one column in the adjacent pair comprises cells storing W− values, wherein adjacent cells in the adjacent pair store a differential weight, W, according to the formula W=(W+)−(W−). In another embodiment, an analog neural memory system comprises a first array of non-volatile memory cells storing W+ values and a second array storing W− values.

A multi-bit cell includes: a memory storing a weight resistance corresponding to a multi-bit weight; a current source configured to apply a current to the memory to generate a weight voltage from the weight resistance; a plurality of multiplexers connected to each other in parallel and connected to the memory in series, each of the multiplexers being configured to output one signal of the weight voltage and a first fixed voltage based on a multi-bit input; and a plurality of capacitors connected to the plurality of multiplexers, respectively, each of the capacitors being configured to store a respective weight capacitance, and to generate charge data by performing an operation on the outputted signal and the weight capacitance.

An N-channel interleaved Analog-to-Digital Converter (ADC) has a variable delay added to each ADC's input sampling clock. The variable delays are each programmed by a Successive-Approximation-Register (SAR) during calibration to minimize timing skews between channels. An auto-correlator generates a sign of a correlation error for a pair of ADC digital outputs. SAR bits are tested with the correlation sign bit determining when to add or subtract SAR bits. First all pairs are calibrated in a first level of a binary tree of mux-correlators. Then skews between remote pairs and groups are calibrated in upper levels of the binary tree using auto-correlators with inputs muxed from groups of ADC outputs input to the binary tree of mux-correlators. The binary tree of mux-correlators can include bypasses for odd and non-binary values of N. Sampling clock and component timing skews are reduced to one LSB among both adjacent channels and remote channels.

A conversion circuit that performs analog-to-digital conversion is described. During operation, the conversion circuit receives an input signal. Then, the conversion circuit performs analog-to-digital conversion and provides a quantized output corresponding to the input signal based at least in part on a first power-supply voltage and a second power-supply voltage of the conversion circuit. For example, the quantized output may be based at least in part on a comparison of the input signal to the first power-supply voltage and the second power-supply voltage. Moreover, the first power-supply voltage and the second power-supply voltage may specify a full-scale range of the conversion circuit. When the full-scale range exceeds a second full-scale range associated with reference voltages that are other than the first power-supply voltage and the second power-supply voltage, the quantized output may correspond to a larger number of bits than when the full-scale range equals the second full-scale range.