synthesizable AM2901 and family bit slice models?

[Mike Kopp]

We are looking at redesigning a PCB with a 16-bit processor built out
of bit slice components. The components used are the AM2901, AM2902,
AM2911, and AM2914. We would put all of the design into an FPGA.

I have been able to find a commercial vendor for the AM2901 (Cast,
Inc.) but they don't have the others. I was also able to find a
synthesizable AM2901 model (I don't remember the URL but it was on one
of the links from the Free Model Foundry,
http://www.freemodelfoundry.com/). This is old (predates std_logic
and I'm not sure about copyright.

Does anyone know of any sources for these models? A couple are
complicated enough to be not terribly difficult but not trivial
either.

 

[Jonathan Bromley]

We are looking at redesigning a PCB with a 16-bit processor built out
of bit slice components. The components used are the AM2901, AM2902,
AM2911, and AM2914. We would put all of the design into an FPGA.

Sorry, I can't help you (but that certainly doesn't mean that such
things don't exist). However.... IIRC the 29xx family weren't
outrageously fast by today's standards - 25MHz-ish??? - and
though it pains me to say it, I can't help wondering whether
it might be easier to write an emulator in pure software,
running on a fast modern general-purpose machine.

Or, since you're in academia and therefore presumably
not hidebound by strange requirements to preserve
the original machine code, it might be easier to re-write
the software to run on a modern platform.

I'm sure you've thought about those possibilities already,
but it is not a good idea to underestimate the difficulty of
making any non-trivial CPU design work 100% right,
particularly if you're trying to preserve exact backwards
compatibility. A software-driven solution may prove to
be less work, and more reliable.
--
Jonathan Bromley, Consultant

DOULOS - Developing Design Know-how
VHDL * Verilog * SystemC * e * Perl * Tcl/Tk * Project Services

Doulos Ltd., 22 Market Place, Ringwood, BH24 1AW, UK
(e-mail address removed)
http://www.MYCOMPANY.com

The contents of this message may contain personal views which
are not the views of Doulos Ltd., unless specifically stated.

 

[Gerhard Hoffmann]

However.... IIRC the 29xx family weren't
outrageously fast by today's standards - 25MHz-ish??? - and
though it pains me to say it, I can't help wondering whether

They were pretty fast for their time in 5 um LS-TTL and would
gain a lot from 90 nm technology. The chip family was documented
very well at the register transfer level and converting this to a
VHDL netlist requires some meticulous work, but is
no intellectual challenge.

Some 2901s + 2902 will collapse to a single DSP48 block
in a Virtex4, plus a few gates. The chips were MSI only
and the IP was in the microcode ROMs. Make sure that you
can read these.

AMD had an excellent booklet series on the 2900 family. The
title was something like "Build your own microprocessor, today".
Described every little bit

but it is not a good idea to underestimate the difficulty of
making any non-trivial CPU design work 100% right,
particularly if you're trying to preserve exact backwards
compatibility. A software-driven solution may prove to
be less work, and more reliable.

It could easily be that 29xx+microcode is better documented
than the resulting machine. And if it's something like a
disk or graphics controller, then there won't be anything like a
conventional CPU, just hardware + microcode.

Gerhard

 

[burn.sir]

AMD had an excellent booklet series on the 2900 family. The
title was something like "Build your own microprocessor, today".
Described every little bit

I think you are referring to "Build a Microcomputer" by AMD.

Looks like someone brave soul saved this book from extinction:
http://buzbee.net/bitslice/

[note: the PDF files are huge, and the site is slow]

Also, bitsavers.org and wikipedia can point you to the complete
datahseets





- Bruns

 

[Eric Smith]

However.... IIRC the 29xx family weren't
outrageously fast by today's standards - 25MHz-ish??? - and

They were pretty fast for their time in 5 um LS-TTL and would

The fast ones were actually ECL internally, with built-in level
conversion on the IO. IIRC, it was a process developed by National
Semiconductor for their IDM2900A (and -B, -C) family, though the
other vendors including AMD followed suit.