Abstract: According to one implementation of the present disclosure, a method for power management is disclosed. The method includes: computing, by a central processing unit, software instructions of a software workload in an active-mode operation corresponding to a first operating point on a performance curve of a performance mode; transitioning from instances of the active-mode operation to instances of standby-mode operation of the CPU, and recording, by a time tracking element, each of a plurality of standby entry data points; transitioning from the instances of the standby-mode operation to the instances of the active-mode operation of the CPU, and recording, by the time tracking element, each of a plurality of standby exit data points; and determining a second operating point on the performance curve of the performance mode based on the recorded standby entry data points and the recorded standby exit data points.

Abstract: In a particular implementation, a method includes: providing a first voltage to a word-line coupled to a first transistor device; providing a second voltage to a bit-line coupled to the first transistor device; providing a third voltage to a source-line coupled between a programmable resistive device and a voltage control element. Also, the first transistor device is coupled to the programmable resistive device and the voltage control element, where the programmable resistive device is configured to replace a first data value by writing a second data value in the programmable resistive device. Moreover, in response to a voltage difference across the programmable resistive device exceeding a particular threshold, limiting the voltage difference by one of reducing the second voltage on the bit-line or increasing the third voltage on the source-line.

Abstract: A method and architecture for performing multiply-accumulate operations in a neural network is disclosed. The architecture includes a crossbar having a plurality of non-volatile memory elements. A plurality of input activations is applied to the crossbar, which are then summed by binary weight encoding a plurality of the non-volatile memory elements to connect the input activations to weight values. At least one of the plurality of non-volatile memory elements is then precision programmed.

Abstract: A multiply-accumulate method and architecture are disclosed. The architecture includes a plurality of networks of non-volatile memory elements arranged in tiled columns. Logic digitally modulates the equivalent conductance of individual networks among the plurality of networks to map the equivalent conductance of each individual network to a single weight within the neural network. A first partial selection of weights within the neural network is mapped into the equivalent conductances of the networks in the columns to enable the computation of multiply-and-accumulate operations by mixed-signal computation. The logic updates the mappings to select a second partial selection of weights to compute additional multiply-and-accumulate operations and repeats the mapping and computation operations until all computations for the neural network are completed.