On the horizon: Fast chips quickly: It's done using silicon compilers. Though they've built some significant circuits, their mare ambitious roles are still in the future

Imagine the following scene. A physicist in need of some special signal-processing chips for a particle-physics experiment goes to a computer terminal and, over a network, accesses a silicon compiler. The physicist specifies the functions the chip is to perform - perhaps data selection, followed by a series of complex transformations, ending with conversion to a standard data format for further processing and analysis (as is done by very large-scale computers in present-day particle physics). The compiler verifies that specified operations applied to imputs with the bit width chosen will give rise to the correct-size outputs. catches a logic error in the data-selection specifications by checking them against the expected properties of the outputs, checks that the algorithm specified by the physicist will work efficiently on silicon, and goes off to compile the chip. A day later, the physicist is informed - probably by electronic mail over the same network used to access the compiler - that the circuit has been designed; it will occupy a 400-by-500-mil area using 1-micrometer design rules. The note adds that it has been placed into the fabrication queue at a nearby silicon foundry and will be ready in four weeks.