Micro Code

MicroCode is a means of implementing a hardware CPU with some assistance from software rather than purely in hardware; typically a user cannot tell whether microcoding was used in implementing a CPU (although there have been some user-modifiable microcoded CPUs).

There is a range of kinds of microcode ranging from "vertical" to "horizontal".

Vertical microcoding means that the underlying CPU is a straightforward one-thing-at-a-time instruction set roughly similar to a RISC processor like a Sparc.

Horizontal microcoding means that the underlying CPU presents many fields per instruction, which each individually control some part of the ALU directly. So one 1-bit field might enable output from register 12 while another 1-bit field enabled input to the AND operation.

This approach gives very precise control over the workings of the ALU, with the ability to specify quite a few things in parallel, but consumes a large number of bits per micro-instruction, since the control fields, in extreme cases, are all just on-off rather than multiplexed.

In the middle of the spectrum, a microcoded CPU might have instructions with only a moderate number of control fields, each multiplexed (e.g. a single 4-bit field might specify one of 16 registers to use as output), and hence take up less space than a purely horizontal microcoded engine, but still more than a highly vertical one.

All other things being equal, a purely vertical microcoded CPU will be slow, and implemented only for prototyping or compatibility, while a highly horizontal microcoded CPU is potentially very fast, but can consume a surprisingly large amount of expensive (in terms of real estate) microcode store.

Microcoding seemed to fall out of favor for some years with the advent of microprocessors, but eventually came into use again as an implementation technique for microprocessor cores.

-- DougMerritt

(although there have been some user-modifiable microcoded CPUs)

also known as "WritableInstructionSetComputers" (WISC).

(moved to page WritableInstructionSetComputer)

What about the TransMeta processors?

Given the market niches they eventually decided to target, changing microcode dynamically isn't part of their business plan. I had the impression that it's all stored on-chip, perhaps in factory-masked ROM, rather than being read from PROM off-chip, let alone from RAM, but I'm not 100% sure.

If their market niche changed, they could easily make a RAM-swappable microcoded version of their cpu...

Of course, these days even things like the Pentium use microcode to drive a RISC-like internal core. Once they got beyond a certain level of complexity, the alternative was insanity. :-)