it looks like the feedthrough path is faster than a normal SRAM read, so we actually get hold time violations at 1.8V / 10 MHz