Re: Silicon Machining

> From: "Kirk W. Fraser" <kirk@...>
> Has anyone considered making their own desktop chip foundry? I read
> that Charles Moore, inventor of Forth got into the National Bureau of
> Standards integrated circuit cell definition library and designed his
> own CPU. I've also read the pamphlet published by Lindsay books on
> making your own mirrors in a vacuum jar. It seemed to me there should
> be some way to combine the two ideas with an external laser under XY
> control to replace some of the expensive chemicals used in industry to
> build integrated circuits in a machine that is no bigger than a
> mill-lathe.
>
> The laser processing would be slower and may not have as fine resolution
> as industry now uses but it could be great for making a few chips to
> build a non-Intel PC and other hard to get devices or repair something.
> Any information to further this like how to access the cell definition
> library, or what size laser is needed to zap away the right parts of
> chips and coating materials without drilling through the whole unit?
> What minimum chemicals are needed to make the different layers?

There are a couple of problems with his idea. First, you are not likely
to be able to make a CPU, due to the cleanliness problems. You could
probably make a flip-flop, or somewhat more complex circuit, but when
you get over 1000 transistors, you really have to have a well-controlled
process, and a cleanroom to have any chance to get the whole chip
working.

2nd, modern chips are made with sub-micron feature sizes, so any
source of red light is too long a wavelength. You would need to go
to blue light to get near the feature size for even common SSI-MSI
chips. And, blue lasers are a LOT more expensive than red.
(This may change soon, people are hard at work on blue lasers, and
making extraordinary progress on them.)

You'd need a VERY precise mechanism for steering the laser beam.
I doubt you'd have much success burning layers off the chip with the
laser. The thermal damage would almost certainly wreck the silicon.

The chemicals needed are not only to etch the silicon and aluminum,
but also to 'dope' the silicon with charge-carrying ions. These are
usually deposited by chemical action at elevated temperatures, with
reactive gases such as diBorane, Phosphine, and Arsenine. I think
all of these are extremely hazardous, and a whiff will destroy your
lungs. In semiconductor fab lines, they have elaborate 'burn boxes'
that surround all the piping for these gases with a metal shield, and
air is sucked through the boxes to burners that destroy any trace
of these gases. That is in addition to sensors for the reactive gases.
Steam is used to create insulating (glass) films on Silicon at 900+
Celcius! Aluminum is eched with Phosphoric acid, Silicon is
etched with Hydrofluoric acid, which is VERY hazardous stuff.

I think it would be possible for a serious experimenter to actually
make an IC, and I believe some science fair winners have done
so. It took them several years to get the processes developed
and under predictable control. And, the IC's they were making
had, maybe 4-10 transistors, enough for one gate or maybe
a simple flip-flop or a simple op-amp.

But, trying to make a CPU or other complex LSI chip is going to make
you very frustrated and very broke! If you really are serious
about this, get in touch with a local university with an Electrical
Engineering department (or possibly, Computer Science) and
see if they are involved in the MOSIS project, which allows
students and researchers to throw a few wafers a month through
commercial fab lines. They put, perhaps, a hundred different
designs on a single mask set and run a couple of wafers through
the fab line, dice it up and send the chips back to the experimenters.
These chips are made on state of the art process technologies,
and so reasonable performance (not the fastest available) can
be obtained from them.

If you really want to design a CPU, or replace obsolete or
unavailable chips, why not use Xilinx programmable gate
arrays or CPLD's? You could definitely build a complex
CPU on one of these, and the upper end of the Xilinx 9500
(in circuit reprogrammable) line can run up to 100 MHz or
so, and Actel has just announced a FPGA that can run at
640 MHz! They have 100,000 usable gate parts in this
line now, and will be making 1 MILLION usable gate parts
next year or so!

Jon