34. Machine learning accelerators
Design data paths for matrix multiplication, convolution, reduced-precision arithmetic, and tensor operations. Compare CPUs, GPUs, neural processing units, systolic arrays, and the effects of sparsity and memory bandwidth.