A multiply-accumulate successive approximation (MASAR) column is provided. The MASAR column includes a plurality of MASAR cells, each including a multiplier configured to perform digital multiplication between an input activation received to an input and an operand to compute a result, and a unit capacitor configured to store the result as analog charge. The MASAR column further includes digital logic configured to perform analog summation of the analog charge of the unit capacitors of the plurality of MASAR cells to determine a digital output of the multiplication.

A multiply-accumulate successive approximation (MASAR) column is provided. The MASAR column includes a plurality of MASAR cells, each including a multiplier configured to perform digital multiplication between an input activation received to an input and an operand to compute a result, and a unit capacitor configured to store the result as analog charge. The MASAR column further includes digital logic configured to perform analog summation of the analog charge of the unit capacitors of the plurality of MASAR cells to determine a digital output of the multiplication.

A hybrid floating-point arithmetic processor includes a scheduler, a hybrid register file, and a hybrid arithmetic operation circuit. The scheduler has an input for receiving floating-point instructions, and an output for providing decoded register numbers in response to the floating-point instructions. The hybrid register file is coupled to the scheduler and contains circuitry for storing a plurality of floating-point numbers each represented by a digital sign bit, a digital exponent, and an analog mantissa. The hybrid register file has an output for providing selected ones of the plurality of floating-point numbers in response to the decoded register numbers. The hybrid arithmetic operation circuit is coupled to the scheduler and to the hybrid register file, for performing a hybrid arithmetic operation between two floating-point numbers selected by the scheduler and providing a hybrid result represented by a result digital sign bit, a result digital exponent, and a result analog mantissa.

Examples disclosed herein relate to averaging modules. For example, a method may include obtaining, by an analog-to-digital converter (ADC), a plurality of samples of an input analog signal. The method may also include determining, by an averaging module, a sum value of the plurality of samples, such that a set of most significant bits of the sum value represent an average value of the plurality of samples. The method may further include determining the average value based at least on the set of most significant bits, and outputting the average value to a bus coupled to a memory module.

Examples disclosed herein relate to averaging modules. For example, a method may include obtaining, by an analog-to-digital converter (ADC), a plurality of samples of an input analog signal. The method may also include determining, by an averaging module, a sum value of the plurality of samples, such that a set of most significant bits of the sum value represent an average value of the plurality of samples. The method may further include determining the average value based at least on the set of most significant bits, and outputting the average value to a bus coupled to a memory module.

An integrated circuit and method are provided for performing weighted sum computations. The circuit includes: a plurality of current generators interconnected and arranged into pairs, a positive summation node, a negative summation node, and an input generation circuit. For each pair of current generators, the control terminal of each element is electrically connected to an input node. One of the current generators has its drain connected to the positive summation node while the other current generation element has its drain connected to the negative summation node. The remaining terminals on both current generators are connected to a reference, which may be shared. Each pair of current generator source predetermined amounts of current onto the two summation nodes when the following conditions occur: the input node is at an activation voltage, and the two summation nodes are at a predetermined target voltage.

A hybrid floating-point arithmetic processor includes a scheduler, a hybrid register file, and a hybrid arithmetic operation circuit. The scheduler has an input for receiving floating-point instructions, and an output for providing decoded register numbers in response to the floating-point instructions. The hybrid register file is coupled to the scheduler and contains circuitry for storing a plurality of floating-point numbers each represented by a digital sign bit, a digital exponent, and an analog mantissa. The hybrid register file has an output for providing selected ones of the plurality of floating-point numbers in response to the decoded register numbers. The hybrid arithmetic operation circuit is coupled to the scheduler and to the hybrid register file, for performing a hybrid arithmetic operation between two floating-point numbers selected by the scheduler and providing a hybrid result represented by a result digital sign bit, a result digital exponent, and a result analog mantissa.

A circuit for fast matrix-vector multiplication and a method for constructing that circuit are provided, comprising processing a matrix to obtain a pair matrix, which is then used to construct a circuit.

A circuit for fast matrix-vector multiplication and a method for constructing that circuit are provided, comprising processing a matrix to obtain a pair matrix, which is then used to construct a circuit.

In an example, a method comprises receiving a first matrix of values to be mapped to a resistive memory array, wherein each value in the matrix is to be represented as a resistance of a resistive memory element. An outlying value may be identified in the first matrix. At least one value of a portion of the first matrix containing the outlying value may be substituted with at least one substitute value to form a substituted first matrix.