A 16-bit Floating-Point Near-SRAM Architecture for Low-power Sparse Matrix-Vector Multiplication | IEEE Conference Publication | IEEE Xplore