Adding big blocks of SRAM to collections of AI tensor engines, or better still, a waferscale collection of such engines, ...