Novel Approximate Multi-precision Bfloat16 multiplier

Improving the efficiency of Approximate Compute Units for AI on Edge via algorithmic optimizations.