Optical computing is right around the corner, and has been for decades. Here's a brief survey of the field from 1997 [1]
Sigh. My honours dissertation was in support of part of the work on one of the MQW devices at Heriot-Watt, as mentioned in the NS article. It was really cute, it would multiply a vector by a 32X32 matrix per clock cycle, and was supposed to be part of an electro-optical neural-net. That didn't go anywhere, as far as I know, but the "smart pixel" idea lives on. --KeithBraithwaite