How hard is to build a processor?

Fri Feb 26, 2010 9:22 pm   

Don McKenzie wrote:

In the early 80's it was common to build "custom" processors
out of 2900 bit-slice components. There was an excellent text
(and some good AMD appnotes) devoted entirely to this (Mick 'n'
Brick? yellow dust jacket).

In grade school, I build a combination (burglar) alarm using
(bistable?) inter-latching relays for the code store. With
a fire department klaxon as the annunciator (you *really*
didn't want to get the combination wrong! :> )

In high school, I built a two-player (offense + defense) football
(left-pondian football, that is :> ) game out of analog computers
(integrators, adders), DTL and VOM's (to display: "down", field
position, yards gained/lost on the play and yards 'til first down)
but that just ran at "DC". It was also quite large (4' x 8'
sheet of plywood to hold all the bits) and, thus, impractical
to preserve.

Many years ago, I built a "digital clock" out of relays. But,
it was very noisey and cost a fortune to keep replacing the
incandescent lamps used in the "7 segment" displays.

Now, I am much more fascinated by electro-mechanical *mechanisms*.
I have been working on a kinetic "sculpture" to act as a timepiece
in the back yard. A tribute to Rube Goldberg -- with the exception
that it must run *continuously* (most of his contraptions were
"one-shot" devices). But, in order to keep *good* time, I need
to "close the loop". Doing so without being noticed means
using some "non-discrete" device that you can control. I.e.,
something like a liquid whose rate of flow can be varied without
a critical observer being able to *easily* determine that this
is happening. Living in the DSw poses a problem using water as
it evaporates too fast (replenishing it from the domestic water
supply would be "cheating" :< ). I also need to locate some
larger solar panels so the device has no connection to the electric
utility.

I would also like to build a "Jetson's" style doorbell (though
programmable) to replace the electronic version I made some years
ago. But, apparently, the design of tubular bells is more art
than science (and, mistakes can be costly). So, I have a lot
more research to do. :<

"Toys" <grin>

Fri Feb 26, 2010 10:18 pm   

Joe Pfeiffer wrote:

Hah! My leg's been pulled. and to think I'd just finished a year on
1401, 705 code! I'd thought 'cadet' was a lab project, not a for-real
machine. I've even read a 1620 manual, but never got to write for one.

-- Bill

Fri Feb 26, 2010 10:36 pm   

"Ian Bell" <ruffrecords_at_yahoo.com> wrote in message
news:hm8aia$ajc$2_at_localhost.localdomain...

@Ian, who did you work for in 73 ???

Fri Feb 26, 2010 10:39 pm   

"Vladimir Vassilevsky" <nospam_at_nowhere.com> wrote in message
news:DLadnf-Fi6vCsRXWnZ2dnUVZ_gednZ2d_at_giganews.com...


Don McKenzie wrote:


It is trivial to build a processor. This type of project is a semester
work for a student. It is hard to make commercially viable processor,
though.

VLV
But you have to add to that all the software to make it do something....
starting with a boot loader...

Fri Feb 26, 2010 11:02 pm   

TTman wrote:

.... and there's the rub! :>

I designed a processor some years ago. A friend was responsible
for writing the code for it.

*Nothing* worked! :< This was completely unexpected as we were
both very competent in our individual responsibilities.

We soon realized that I had designed the instruction set expecting
"word" addresses (memory was 16b wide and only accessible *as* 16-bit
words -- hence it seemed *obvious* that addresses would be of "words")
whereas he had assumed *byte* addresses. :< Simple fix. Took
all of the drama out of the event! ;-)

I've seen other silly issues like this confound the initial
startup of custom processors: e.g., confusion over which
way the stack grows, whether the SP points to the last *used*
location on the stack or the next *available*, etc. They are
almost always "fun" problems to solve as they usually are
easy to find and have dramatic consequences once found!

Sat Feb 27, 2010 1:09 am   

On Fri, 26 Feb 2010 14:18:14 -0700, Bill Cooke
<bcooke_at_cookedata.com> wrote:


I can remember hearing that phrase from time to time when I
worked on the 1620. It was a fun machine. I used to swap
out the colored bezels on the control panel just to tease.

Jon

Sat Feb 27, 2010 2:33 am   

Jon Kirwan wrote:

I was working at a small shop that had a hand-me-down 1620. They had a simple
Fortran compiler, "PDQ Fortran", but it lacked some features they wanted. On my
own, I wrote a disassembler for the 1620, disassembled the compiler, studied the
code, figured out how to save some code space (memory was limited), then add
some features, basically enhanced write commands and formatting, all with no
external documentation. When the code was almost ready I was so excited I
couldn't sleep, so went into work at 4 AM or such and got it working. Fun days!

--
Thad

Sat Feb 27, 2010 2:39 am   

D Yuniskis <not.going.to.be_at_seen.com> writes:

Yes, Mick and Brick. An absolutely outstanding book on datapaths. and
microprogramming; it was all based on 2900-series, but the concepts
mapped to everything.
--
As we enjoy great advantages from the inventions of others, we should
be glad of an opportunity to serve others by any invention of ours;
and this we should do freely and generously. (Benjamin Franklin)

Sat Feb 27, 2010 4:57 am   

On Feb 26, 3:22 pm, D Yuniskis <not.going.to...@seen.com> wrote:

Yup! I read Mick and Brick cover to cover and learned a lot from it.
I never built anything with bit slice other than on paper. But I have
a lot of respect for those who did. I even used a "high end"
workstation once that was a suped up 68000 made out of bit slice. I
think it had a marketing window of 15 minutes before Moto came out
with a 680xx or something much faster than the 68000.



I have thought about how to make a time piece that is actually
regulated by the flow of water. It would be hard to get this to be
accurate, but I havae some ideas on how to make it fairly good. I am
in the mid-east US, so we normally have lots of rain. I have thought
about ways to make it "self-winding". One is to simply catch rain
from the roof and keep the top reservoir full. Another would be to
use wind power to pump water from the lower reservoir to the top.
That would be doubly cool. It might even allow the clock hands to be
in front of the windmill blades!

But this project is way off in the distance. I have many other things
to do first.

Rick

Sat Feb 27, 2010 6:47 am   

The clock/calendar I hope to build over the next year or so will be
solar. The shadow of a post uniquely determines both date and time, if
you look at both angle and length....
--
As we enjoy great advantages from the inventions of others, we should
be glad of an opportunity to serve others by any invention of ours;
and this we should do freely and generously. (Benjamin Franklin)

Sat Feb 27, 2010 6:49 am   

Hi Rick,

rickman wrote:

AMD published a lot of app notes and manuals that really catered
to the 2900 family of devices. In my Great Databook Purge, they
are among the few items that I kept! (along with Mick 'n' Brick,
of course!)


I plan on cheating: detecting the "displayed time" and using
that in a feedback loop to control the pump speed. It would
probably need to be a terribly overdamped control system
given all the other "cruft" between the pump and the
"display".


Yes, that's my approach. We don't have enough rainfall to
"self wind". I suspect it would be very difficult to keep
enough water in the system to span the gaps between rains!
(or, if you could keep enough water, trying to keep that water
"clean" of algae, etc. over that long of a time period).

We get *lots* of sun so PV seems to be an essential part
of any solution.


Ah, I don't plan on displaying the time in such a "traditional"
format. :> I don't want folks to recognize it as a timepiece
unless they *know* how to "read" it. Instead, it will just
look like a kinematic sculpture...


<grin> Yup. In my case, the problem is figuring out what
*will* work "on paper" before investing lots of time building
something that just turns out to be a nonfunctional eyesore.

Sat Feb 27, 2010 7:06 am   

Hi Joe,

Joe Pfeiffer wrote:

Hmmm... is that (really) true? Or, don't you end up with
*two* date,times for each angle,length? E.g., won't the
angle,length be the same for HH:MM on the day before and
after the Summer Solstice? Or, close enough to make it
near impossible to differentiate? (dunno, I find thinking
in 3D on astronomical scales difficult :> )

Like me, at least you'll have plenty of Sun to play with! (NM)

Sat Feb 27, 2010 5:11 pm   

D Yuniskis <not.going.to.be_at_seen.com> writes:


Yes, I should have said "just about unique." I wouldn't be at all
surprised to find out the variation with date won't be possible to
distinguish more accurately than a couple of days, too.


I like living down here!
--
As we enjoy great advantages from the inventions of others, we should
be glad of an opportunity to serve others by any invention of ours;
and this we should do freely and generously. (Benjamin Franklin)

Sat Feb 27, 2010 9:05 pm   

On Sat, 27 Feb 2010 12:11:32 -0800
John Nagle <nagle_at_animats.com> wrote:


Muscle wire's a pain to work with. In free air you wind up needing to
run a pretty decent amount of current to replace the heat you lose to
natural convection. Even more so if you've got a breeze.

How about the pendulum drives a small DC motor/generator? That lets
you both monitor the frequency and give it little kicks to keep it
on course. With good bearings it shouldn't take much energy to
overcome the losses.

--
Rob Gaddi, Highland Technology
Email address is currently out of order

Sat Feb 27, 2010 9:11 pm   

(Changed title, trimmed irrelevant groups.)

D Yuniskis wrote:


Look into shape-memory alloy wire. See

"http://www.amazon.com/gp/product/B000FMUEGI"

Run a current through that, and it contracts slightly. Not much.
Then obtain or build a pendulum clock, and use the shape-memory
wire to change the supported length of the pendulum. You'll
only get 1-2% change, but that's all you need to fine-tune
the thing.

John Nagle