Adaptive digital-to-analog converter range optimization

A system comprising:

• a processor, and
• a memory storing a plurality of instructions for configuring the processor to:
• obtain an input distribution of data input values to be received at an analog-in memory computing (AIMC) device combining a plurality of computing elements in a crossbar arrangement to form a neural network model;
• select an initial maximum data input value range from said input distribution to be received at the AIMC device;
• perform neural network model operations using quantized input data values mapped from data values within said initial maximum data input value range and obtain an initial model output value;
• determine a performance metric based on said obtained initial model output value;
• and
• tune the maximum data input value range to maximize the performance metric, wherein to tune, said processor is further configured to clip outlier values of the input distribution of data input values to restrict the input distribution range of data input values and increase a signal to noise (SNR) ratio of a model output value when performing the neural network model operations.

A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions readable by a device to cause the device to:

• obtain an input distribution of data input values to be received at an analog-in memory computing (AIMC) device combining a plurality of computing elements in a crossbar arrangement to form a neural network model;
• select an initial maximum data input value range from said input distribution to be received at the AIMC device;
• perform a neural network model operations using quantized input data values mapped from data values within said initial maximum data input value range and obtaining an initial model output value;
• determine a performance metric based on said obtained initial model output value;
• and
• tune the maximum data input value range to maximize the performance metric, wherein to tune, program instructions further cause the device to clip outlier values of the input distribution of data input values to restrict the input distribution range of data input values and increase a signal to noise (SNR) ratio of a model output value when performing the neural network model operations.