The memory hierarchy and caches

In April 1965 Maurice Wilkes of Cambridge published a two-page note in the IEEE Transactions on Electronic Computers. He proposed placing a small, fast memory, which he called a slave, in front of the large main memory, and keeping in it the words the processor had used recently. Most accesses would find their data in the slave; only the rest would pay the full price of main memory.

The word cache came into use a few years later. Today a processor core spends more time and silicon on its caches than on arithmetic, because the gap Wilkes saw has grown to roughly a hundred processor cycles for every trip to DRAM.