Ethan's Retrocomputing Corner


classiccmp: Core Memory Theory and Construction

Date: Mon, 19 Apr 1999 07:45:59 -0700 (PDT)
From: Ethan Dicks 
Subject: Re: How scarce (valuable) is core for the PDP-8?

When I mentioned the chance to buy a 4kW stack for the PDP-8/i for $100...

--- Lawrence LeMay  responded:

> Actually, that's probably a reasonable price.

Foo!

> Core memory boards, probably non-working, have been going for a high price. 

I got sniped for a PDP-11 double-core stack this weekend, backplane included,
that went for $38, no reserve.

> Age and a nice visible setup increase the price. 
 
The core stack for a PDP-8(i|L) is older than much of what's on the
market, but none of the good stuff is visible at all on it.

> Now, I haven't seen the memory in question. but the pdp8/e core
> memory i've seen is all covered by a clear plastic shield. This
> increases its value as a display piece, as you can easily see
> all the core, and its all protected.

It's hard to describe the arrangement, but the core plane in question
here is a block with two edge-connectors on either side, "dual-height"
as they say, but it's much thicker - let's try bad ASCII art to illustrate...


  ########  ########
  xxxxxxxxxxxxxxxxxx == ######## ########
  ########  ########    ######## ########

  core planes            paddle-boards with wire harness

The outside of the core plane part is covered in a "diode matrix", with
a wad of twisted-pair wires that go off to paddle-boards, one for the
sense bits, one for the inhibit bits.  The address lines come up the diode
boards, the data comes up and down the paddle-boards. 

There are several PCBs with core in the core stack, 4-bits per layer with
an optional parity layer that has one pad of bits and three pads of core-less
X-Y wires.  None of this is visible when the plane is assembled, and it's
soldered together with lines of wires going up and down the planes. 

> Of course, in order to use the core on a pdp8/? you would need
> a couple of support boards in addition to the core plane board
> itself. I would say that just the core plane, being of a nice
> size, and being very good 'visually' to display, and somewhat
> because its a PDP8 series board (nostalgia value), that its
> probably worth $100 all by itself. If it comes with the 2 support
> boards and the top connector things at that price, then i'd say
> its a bargain.

You are thinking of newer hardware.  The pre-OMNIBUS 8's have a wad of
individual, single-height cards that contain the sense-amps and the inhibit
drivers.  I have a pile of them from an -8/L that someone else had already
begun to strip for parts before I bought it (it also happens to contain the
only DEC lock that does *not* use the XX2247 key).  I'm not worried about
the analog stuff... I need the core.

Of course, as Allison pointed out, I could always stick in a lump of battery-
backed static RAM.  I was contemplating building a wiring harness to adapt
an RX8E on the back of either an -8/L (which has 8kW of core out of 12kW in
an expansion cabinet) or on the -8/i.  I would use berg connector pins to
stick the wires on the back side of the backplane (to avoid soldering, of
course; but worst case, I just wire-wrap on a connector or two and use
sheilded ribbon to move the signals around.

The joys of restoration in a market of scarcity. :-P

-ethan

Date: Mon, 19 Apr 1999 09:23:49 -0700
Reply-to: classiccmp@u.washington.edu
From: Chuck McManis 
Subject: Re: How scarce (valuable) is core for the PDP-8?

At 04:58 AM 4/19/99 -0700, Ethan Dicks wrote:
> So... for those people who have been trying to get core over the past year
> or two, what's it costing?  I'm trying to decide if I want to grab this
> stack to put into my -8/i and bring it up to 8kW, an entirely optional
> project (I have all the other parts I would need for the upgrade from a
> PDP-8/L that I got in 1982 that was sold as parts-only, bad core, and most
> of the I/O and part of the CPU missing).

As Allison pointed out, its called a "stack" for a reason as I've
discovered over the last 20 months.

Core memory consists of three major elements, the select logic, the planes
themselves, and a set of current source/sense amplifiers that are used to
drive it. Depending on the design these can be somewhat generic or not at
all generic. The absolute best value is to get a complete set of core plane
boards that came out of the DEC factory working. The second best is to get
a set that DEC had refurbished, and everything else loses value from there.

When I joined this list, I did so because I was looking for some memory for
a PDP-8/M that I had. The system was complete except for memory. Later, I
discovered that 4K wasn't sufficient really to run much more than the paper
tape stuff so I really needed 8K.

I have searched fairly widely, both through the internet and through the
"unofficial" DEC channels. This is what I've discovered.

DEC spares still carries core for the 8/e/f/m, 4k is $1,725 per set.
Ideally you need 12K for OS/8 and that means you really want an 8/E dual
backplane or an expansion box.

Many of the DEC "wholesalers" have a very small selection of OMNIBUS
boards, I discovered Mitch over at Keyways and bought from him a G104,
G227, and H220 board (one 4K stack), but they were not a matched set.
(total cost $275) While Mitch would have been happy to take them back, I
chose to keep them on the assumption that getting a complete set working
would be easier than finding a complete set.

Every now and then one of the wholesalers has a complete PDP-8 for sale. I
found one in LA for $800. I had a friend check it out and it was a dead
8/f. Memory was third party (DATARAM) and couldn't be verified. Pass.

Two months ago I found an 8/M at auction at SRI with a 8K stack and a 4K
stack in it. I bid $206 and won it. Amazingly it actually worked so I
either managed to get memory for my non-working 8/M or I still have an 8/M
with no memory and an extra 8/M. :-)

The bottom line is that real honest to god core that works is pretty hard
to come by. I've never seen any core for the 8/I or 8/L (except the ebay
one, and the 8/L that wsa offered for trade on this list).

There is an excellent theory of operation on core planes in the PDP-8/a
miniprocessor users handbook. Between that, my current probe, and my scope
I believe I can get the dead 4K stack I've got running again. And I'm
currently building a 32KW semiconductor board so that I'll have a backup.

--Chuck

Date: Mon, 19 Apr 1999 20:31:49 -0500 (CDT)
From: Lawrence LeMay 
Subject: Re: Repairing core (was Re: How scarce (valuable) is core for

> > [For the record a 4K core stack in an 8 consists of a G104, G227, and H220
> > card. It was the H220 card that had the actual core "mats" on it. The H220
> > card is often "liberated" from PDP-8's as a trophy/display item.]
> 
> Yes, but... The H220 (or the 8K core plane that I've forgotten the name 
> of) has 8 connectors on it (4 into the bus, 4 to top connectors to the XY 
> and sense/inhibit boards). When in place it's fairly well hidden by those 
> top connector blocks as well. So it's possible the core planes _were_ 
> still in the machine, but they hadn't spotted them. Unlikely, I guess.
> 

My 8K set is H212, G104, G233

My 4K set is H220, G227, G111

Its possible the previous owner mixed up the boards though.

-Lawrence (Whistling "If I only had a PDP-8/e" from the Wizard of OS8) LeMay

Date: Tue, 20 Apr 1999 16:46:24 -0400 (EDT)
From: allisonp@world.std.com
Subject: Re: How scarce (valuable) is core for the PDP-8?

I wager the wire used in the core is finer than #30, my bet is #40.


> I do not believe my engineering and math skills are adequate to design it,
> but given the specs, I sure could _build_ it.  If anyone is considering 
> this, please bore out the threads of the nut (if you use nuts) before
> determining the saturation current.  It will minimize mechanical wear on
> the insulation.  I may just go to the local fastener house and buy a box
> of #2 nuts and build a mat of this just to hang on the wall.  ;-)

Skip all the fancy sillyness.  All I need is s pulse generator that can
generate fast bidirectional pulses with currents to 1-2A and a scope.
from that I can tell what current saturates the core and what it's switch
thresholds are. actually a power supply abnd  resistor and a switch will
do for the pulse generator, though it's more manual.  the rest is easy.  
The array will have to be tuned once built anyway but knowing what the X
and Y currents have to add to +/- 10% is enough to get it going. The 
inhibit current will be the mirror of the half select current.

Oh, the same rig will be a good apprasial of the figure of merit for the
core.

Allison

Date: Tue, 20 Apr 1999 22:27:34 -0400
From: allisonp@world.std.com (Allison J Parent)  | Block address
Subject: Re: How scarce (valuable) is core for the PDP-8?

> generator set up to drive a fast rise time buffer amp (so I can get a
> reasonable amount of current) what else do I need to do to verify that I'v
> successfully "flipped" the device. (My first guess would be a garden
> variety compass)


first you need a waveform that looks like this....

  
  _____|-----|_____________      _______+I
                           |_____|      -I

And put about one turn through the ferrite or steel under test.

Then you'll need a second 1-2 turn loop hooked to the scope.

You'll know the right waveform when the signal starts showing a huge
amplitude difference as switches.

Allison

Date: Wed, 21 Apr 1999 15:41:42 -0400 (EDT)
From: allisonp@world.std.com
Subject: Re: How scarce (valuable) is core for the PDP-8?

> Most of the DEC core planes use a fair number of special-purpose
> transistor arrays (often specially matched) for drivers as well as pulse
> transformers and delay lines.  If you look back at alt.sys.pdp8 archives
> (or its mailing-list twin, PDP8-LOVERS) you'll find some suggestions
> for part substitutions and/or rebuild directions.

If your repairing a existing core that is significant.  If your doing
something small that  just has to work for demo reasons that is not a show
stopper.

> of driving and sensing.  There were at least a few implementations of
> S-100 core memory in the mid-70's, and one of them was a S-100 to
> Unibus translator of sorts.  (Not to be confused with the IMSAI
> S-100/Unibus shared memory module, which is in several of my IMSAI
> price lists from 1977/1978 but which I believe to be vaporware.)

Byte, kilobaud or interfaceage in the late 70s (76 thru 79) I distinctly
remember an articale with some detail.


I have an advantage... 8E printset and a Electronics engineering caseboot
describing several designs from the late 50s including a code converter
and the Lincoln TX2.  There are enough fragments there for me to do
something.

Allison

Date: Sat, 24 Apr 1999 21:23:40 -0700
From: Chuck McManis 
Subject: Building Core, Stage 2

Ok, I'm getting to the point where I can _almost_ start taking pictures of
the waveforms on the oscilloscope. 

Now, if you are wondering, this does have a point. I've got a 4K core stack
for a PDP-8 that I've been loath to get into because I didn't feel I had a
strong enough understanding of the basics in order to figure out just what
the heck was wrong with it, once this project is complete and I understand
core memory, theory and operation, I'll be ready to tackle fixing that stack. 

I've been reading and re-reading the theory section in the PDP-8a manual
and some parts are starting to seep into my consciousness. This is where we
are at.

1) I'm using copper clad board to provide a ground plane "under" the core.
After my last message it became obvious that this was the other half of the
winding "through" the core. 

2) I'm using "magnet" wire (enameled copper wire) for the core wire. I'm
guessing it can take a couple of amps for a short period of time. I'm not
sure that is sufficent to "switch" the #2 nut I'm using but we'll see. 

3) Allison referenced a waveform that looked like this:

           +--+
           |  |
           |  |
 ----------+  +-------------+  +--------- Gnd
                            |  |
                            |  |
                            +--+

Now in the ToO section in the manual it was written that the first pulse
"reads" the core, and the second pulse "writes" the core. A light bulb went
on because of course core is "destructively" read, and has to be rewritten.

4) The diagram in the 8a manual shows the sense wires going through one set
of cores, looping and coming back through another set. The sense wire is
also used as the 'inhibit' wire when one wants to write a zero. This is
done by reversing the X or Y current so that the selected cores don't get
full write current (and thus don't switch.)

5) So this is the plan:
        put three wires through the nut and wire them up as follows:
                                 --
        1/h select ----------\    | |        /----- GND
        1/h select ---------------| |-------------- GND
        sense +    ----------/    | |        \----- sense -

   The scope will display the current on the select lines on channel 1
   the sense lines on channel 2 (floating ground)

6) When the current is sufficient, I should see a pulse on the sense line.
   (This is the hypothesis part, now to apply the scientific method)

We'll see ... if there are obvious errors in the above please let me know!

--Chuck

Date: Sun, 25 Apr 1999 10:15:00 -0400
From: allisonp@world.std.com (Allison J Parent)
Subject: Re: Building Core, Stage 2

> 1) I'm using copper clad board to provide a ground plane "under" the core.
> After my last message it became obvious that this was the other half of th
> winding "through" the core. 

Not required.  Any wire with current passing through it has a field around 
it.  That is all that is required.  More turns only reduce the current 
required to attain a given field.

> 2) I'm using "magnet" wire (enameled copper wire) for the core wire. I'm
> guessing it can take a couple of amps for a short period of time. I'm not
> sure that is sufficent to "switch" the #2 nut I'm using but we'll see. 

The current can be quite high as those pulses will be short (microseconds).

> 3) Allison referenced a waveform that looked like this:

Keep in mind that will switch the core in both directions.  that was 
mostly a test signal for trying materials.

> 4) The diagram in the 8a manual shows the sense wires going through one se
> of cores, looping and coming back through another set. The sense wire is
> also used as the 'inhibit' wire when one wants to write a zero. This is
> done by reversing the X or Y current so that the selected cores don't get
> full write current (and thus don't switch.)

Yep the two matic selects and the sense/inhibit wire.  Some planes used four 
wires with the inhibit being seperate.

> 5) So this is the plan:
>        put three wires through the nut and wire them up as follows:
>                                  --
>        1/h select ----------\    | |        /----- GND
>        1/h select ---------------| |-------------- GND
>        sense +    ----------/    | |        \----- sense -
 
For testing a "core" only two are needed.  makes life simpler.

>    The scope will display the current on the select lines on channel 1
>    the sense lines on channel 2 (floating ground)

Good!

> 6) When the current is sufficient, I should see a pulse on the sense line.
>    (This is the hypothesis part, now to apply the scientific method)

Actually even if the current is low there will be a pulse coincident with 
the drive pulse (transformer action).  However when the core switches
the waveform will not corospond to a 1:1 transformer and also it will be 
shifted in time.  If the core is not suitable the signal may not be 
visible.

Allison

Date: Mon, 26 Apr 1999 14:51:16 -0700
From: Chuck McManis   | Block address
Subject: Cores, light bulb goes on ...

Core memory, the saga continues. 

Ok after spending much of the weekend trying to get my core driver working,
and feeling really good about having all the parts I needed in my spares
box, I realized I had built half of an h-bridge. Since I had already
designed a really nice MOSFET based h-bridge that could switch 10 - 15 amps
for my robots I slapped my head and went "Doh!" 

Anyway, so I wired up my #2 nut with sense wire to the h-bridge and start
slapping current back and forth through my nuts. :-) Seriously though when
viewed on 'scope it looks something like:

     +--+
     |  |
Ch1 -+  +----+  +-----
             |  |
             +--+
     +--+    
Ch2 -+  +----+  +-----
             +--+

UNTIL you get to about 7.5 amps or so, and then it looks like:

     +--+
     |  |
Ch1 -+  +----+  +-----
             |  |
             +--+
       +--+    
Ch2 ---+  +----+  +-----
               +--+

So I stared at it a bit and the little bulb went on between my ears.

The first pulse "writes" a zero to the core, if the core is either already
magnetized with a zero, or has nothing in it, then the sense line sees the
deta-B (change in magnetic field) and a current is induced in the wire.
When the second pulse comes along it "writes" a one into the core, same
effect on the sense wire. The clever part comes when your "write" a zero to
a core that has a one already written to it.

In this case the current in the wire induces its standard magnetic field,
but that field is _cancelled_ by the field already in the core, thus for a
period of time the sense line sees no change in magnetic flux, and so no
current is induced. Then however the core switches to 0 and the sense wire
sees that change in flux and out comes the induced pulse now delayed from
the initial point by a time controlled by how long it took to saturate the
core and switch it. 

The gap between the pulse start and the sense pulse is used to tell whether
or not the core had a 1 in it. Now in the DEC design what happens after the
read pulse (which is really a "write zeros" pulse, is they take the data
they just read and re write with the write ones pulse. However this time
since the sense lines aren't needed to figure out what the cores had in
them, they use them for "inhibit" currents.

Remembering that the cores were all written to zero by the 'read' when the
write 'ones' pulse comes along you don't want it writing cores that you
want to keep zero (they had zero before). So a current that is in opposite
direction of the write current is sent down the sense lines where zeros
should be kept. The magnetic field generated by these "inhibit" currents
cancel the magnetic field created by the "write ones" current and thus
prevents those cores with a 'zero' in them from being written to '1'.

--------------------------
Cool stuff, now it raises some new questions:
        1) Do you want your pulses to be long enough to switch the
           core exactly, or longer? (eg does writing a zero just cancel
           a one or does it cancel the one and write a zero in its place?)

        2) Why not just gate the write one current pulse? That would save
           on the inhibit current stuff.

        3) What properties of a material make it easier to switch at lower
           currents? I don't want to build a core plane with nuts if I need
           8 amps to switch them.

--Chuck McManis

Date: Mon, 26 Apr 1999 19:19:07 -0400
From: allisonp@world.std.com (Allison J Parent)
Subject: Re: Cores, light bulb goes on ...

> Ok after spending much of the weekend trying to get my core driver working
> and feeling really good about having all the parts I needed in my spares
> box, I realized I had built half of an h-bridge. Since I had already
> designed a really nice MOSFET based h-bridge that could switch 10 - 15 amp
> for my robots I slapped my head and went "Doh!" 

;)  that will do.

> Anyway, so I wired up my #2 nut with sense wire to the h-bridge and start
> slapping current back and forth through my nuts. :-) Seriously though when
> viewed on 'scope it looks something like:
> 
>      +--+
>      |  |
> Ch1 -+  +----+  +-----
>              |  |
>              +--+
>      +--+    
> Ch2 -+  +----+  +-----
>              +--+
 
Transformer action prior to saturation.

> UNTIL you get to about 7.5 amps or so, and then it looks like:

No, you know what nuts don't work...

>      +--+
>      |  |
> Ch1 -+  +----+  +-----
>              |  |
>              +--+
>        +--+    
> Ch2 ---+  +----+  +-----
>                +--+
>
> So I stared at it a bit and the little bulb went on between my ears.
>
> The first pulse "writes" a zero to the core, if the core is either already
> magnetized with a zero, or has nothing in it, then the sense line sees the
> delta-B (change in magnetic field) and a current is induced in the wire.
> When the second pulse comes along it "writes" a one into the core, same
> effect on the sense wire. The clever part comes when your "write" a zero t
> a core that has a one already written to it.
 
Core is one of the few DESTRUCTIVE READ memories.  You have to try and flip 
it to see if it flips, if it does you had a (one or zero) and if didn't
you have a (zero or one). 

Designing circuits the transformer and the "bistable" properties makes for 
interesting logic.

> In this case the current in the wire induces its standard magnetic field,
> but that field is _cancelled_ by the field already in the core, thus for a
> period of time the sense line sees no change in magnetic flux, and so no
> current is induced. Then however the core switches to 0 and the sense wire
> sees that change in flux and out comes the induced pulse now delayed from
> the initial point by a time controlled by how long it took to saturate the
> core and switch it. 

Yep.  Watching that spring snap in interesting on a scope.  Those nuts are 
slow too!  Try a bunch of other materials now that you're set up... 

> The gap between the pulse start and the sense pulse is used to tell whethe
> or not the core had a 1 in it. Now in the DEC design what happens after th
> read pulse (which is really a "write zeros" pulse, is they take the data
> they just read and re write with the write ones pulse. However this time
> since the sense lines aren't needed to figure out what the cores had in
> them, they use them for "inhibit" currents.
 
They used seperate wires (four wire cores) in some cases as it was easier
that all that switching.

> --------------------------
> Cool stuff, now it raises some new questions:
>        1) Do you want your pulses to be long enough to switch the
>           core exactly, or longer? (eg does writing a zero just cancel
>           a one or does it cancel the one and write a zero in its place?)
 
I forget.

>        2) Why not just gate the write one current pulse? That would save
>           on the inhibit current stuff.

well you have to know what you're writing back and to do that you have to 
read it first.  Hence the common write after read cycle on many machines 
from the era of core.  

>        3) What properties of a material make it easier to switch at lower
>           currents? I don't want to build a core plane with nuts if I need
>           8 amps to switch them.

Good magnetic conductors that hold their magnetizm.  Some steels, ferrites, 
cobalt alloys, alnico, a few rare earths.  Try some of those ferrite beads
used for bypassing in RF work.

Allison

Date: Mon, 26 Apr 1999 17:12:30 -0700 (PDT)
From: Dwight Elvey   | Block address
Subject: Re[2]: Cores, light bulb goes on ...

allisonp@world.std.com (Allison J Parent) wrote:
> Transformer action prior to saturation.
> 
> > UNTIL you get to about 7.5 amps or so, and then it looks like:
> 
> No, you know what nuts don't work...

Hi
 Although not done on the original cores, you can always increase
the fields with more turns. It doesn't invalidate the demonstration.
Taking your 7.5 amperes and making ten turns drops it to about
750 ma. It also slows down the circuit by the square of the
number of turns though.

Dwight

Date: Tue, 27 Apr 1999 21:20:50 +1000
From: Huw Davies  
Subject: Re: How scarce (valuable) is core for the PDP-8?

At 08:41 22/04/99 -0400, allisonp@world.std.com wrote:

> However with 256Words of core working building a simple TTL machine around
> it would be of some educational value.  It is however snough space to run
> practical demonstration programs.

There was a construction project in a local electronics mag (Electronics
Australia) around the same time as the original Altair articles describing
how to build a computer closely based on the PDP-8. This system had 256
words of memory. I should dig up the articles and this would probably be a
neat starting point for a CDS (Core Demonstration System). Please note: CDS
is a trademark of HOC2 (Huw's Old Computer Company - cf EMC2) :-)

 Huw Davies                      | e-mail: Huw.Davies@latrobe.edu.au
 Information Technology Services | Phone: +61 3 9479 1550  Fax: +61 3 9479 1999
 La Trobe University             | "If God had wanted soccer played in the
 Melbourne Australia 3083        | air, the sky would be painted green"

Date: Wed, 28 Apr 1999 15:00:35 -0700
From: Chuck McManis 
Subject: Re: Cores, light bulb goes on ...

At 06:40 PM 4/28/99 +0100, Tony wrote:
> Incidentally, I believe that the HP9100B core address driver is very 
> simple. Just a emitter follower (with a low-ish emitter resistor) 
> capacitively coupled to the address wire. This gives the required +ve and 
> -ve spikes when the transistor turns on and off.

That's pretty clever, what size cap?

I = Cdv/dt on the cap would allow for an arbitrary spike.

I'll wire this up tonight. I bought some ferrite cores from a local surplus
shop on the reccomendation of one of the core manufacturers and they have
two very nice attributes:
        1) They are all consistent magnetically.
        2) Their magnetic properties are well characterized so I can
           relate theory to actual practice with fewer variables.

Note you those of you in school or headed there, even the most obscure
classes come in handy so PAY ATTENTION. :-)

--Chuck

Date: Thu, 29 Apr 1999 00:22:07 +0100 (BST)
From: ard@p850ug1.demon.co.uk (Tony Duell) 
Subject: Re: Cores, light bulb goes on ...
 
> At 06:40 PM 4/28/99 +0100, Tony wrote:
> > Incidentally, I believe that the HP9100B core address driver is very 
> > simple. Just a emitter follower (with a low-ish emitter resistor) 
> > capacitively coupled to the address wire. This gives the required +ve and 
> > -ve spikes when the transistor turns on and off.
> 
> That's pretty clever, what size cap?

I slightly mis-remembered it, but I've looked in my notes again, and it's 
something like this : 

        ---------------- +ve supply (via decoder tree)
       /
     |<
     |   PNP transistor
     |\
       |       ||
       +-------||------- Core Address line
       |       ||  
       V       27nF
       - 
       |
     -----
Ctrl-|   |  Current Sink 1
     -----
       |
       +--------------- Common return of core address lines
       |
     -----
     |   | Current Sink 2
     -----
       |
       +---------------------------- -ve supply

The transistor base comes from the address decoder (and is gated by a timing
pulse. Current sink 1 (discharge current) is also controlled by a timing 
pulse. Note that the current sinks are common to all X or Y wires - each 
driver consists of a transistor, diode and capacitor (and the decoder 
output to operate the transistor base)

It appears that the 'charging current' flows through the PNP transistor, 
the capacitor, core address line and back through sink 2. The 
'discharging current' flows through the cap, the core address line (in 
the other direction), the diode and sink 1.

If I had a current probe (one of the few useful pieces of test gear that 
I don't have...) I'd clip it on the test loops on the address PCB (one 
for the X direction, one for the Y direction) and see what the waveforms 
look like.

-tony

Date: Thu, 6 May 1999 22:43:28 -0400
From: allisonp@world.std.com (Allison J Parent)
Subject: core... more

Well thanks to John D.  I have a bunch of cores from the early 60s.

They are roughly 50mils x 11 mils x 15mils.  Cross section of the doughnut 
is rectangular at 15x11 mils.  The reason for such rough measurements is my 
vernier is only good to .001" and I'd need somthing fancier to be more 
accurate.  By eyeball the 8e cores are smaller!  Based on several articles 
I've read theses will need about 400->600ma to switch and will do so in 
under 4uS.  I will have to test this in a jig.  Since the hole is ~20mils 
several #40 wires should pass through it easily.

A simple core frame would be 8x8 (64 bits) and use a 4 wire system as 
that simplifies the select, inhibit and sense hardware.  I'd likely go 
with late 70s level TTL and transistors to drive these and to sense the 
outputs I don't know if I'll use transistors (1968 or earlier designs) or 
comparator chips (aka 1540, 710, 711) will be used.  They would also be 
consistant with 1970s technology.  The goal if I can get the time is a 
64x8 or 64x12 memory.  Not large but enough. Why 64?  becuase 1 of 8 
decoders were common even in 1970, and the larger the array the more 
noise from switching.  So 8x8 is manageable.

Allison

Date: Fri, 7 May 1999 09:00:12 -0500
From: "John Dykstra" 
Subject: RE: core... more

On Thursday, May 06, 1999 9:43 PM, allisonp@world.std.com 
[SMTP:allisonp@world.std.com] wrote:
> They are roughly 50mils x 11 mils x 15mils.  Cross section of the
> doughnut is rectangular at 15x11 mils.  The reason for such rough
> measurements is my vernier is only good to .001" and I'd need
> somthing fancier to be more accurate.  By eyeball the 8e cores are
> smaller!

According to Pugh's "IBM's 360 and Early 370 Computers", in 1963 IBM was 
manufacturing both 30x50 mil and 19x32 mil cores (inside diameter x outside 
diameter).  Allison's measurements match up with the larger size.

According to the book, the IBM 7080 and 7094 used the smaller size.  Thus, 
these cores were probably intended for either the 1401, or the 7090 if any of 
these were still in production in 1963.

Does anyone out there have a 1401, or a 1401 core stack?

Something I hadn't read of before:  the 7090's core planes were immersed in
oil to help with heat dissipation.  I wonder whether there was a sticker on
the machine (right underneath "Trained Service Personel Only"):  "Change oil
every six months or 100,000 punched cards".

Another interesting point:  The defect ratio of cores that reached the plane 
wiring stage was 1 out of 8,000.  This sounds incredibly high, but IBM's core 
manufacturing facilities were state of the art.  Imagine the difference
between the process control used then, and current integrated circuit
fabrication techniques!

> A simple core frame would be 8x8 (64 bits) and use a 4 wire system as
> that simplifies the select, inhibit and sense hardware.  I'd likely go
> with late 70s level TTL and transistors to drive these and to sense the
> outputs I don't know if I'll use transistors (1968 or earlier designs)
> or
> comparator chips (aka 1540, 710, 711) will be used.  They would also be
> consistant with 1970s technology.

I'd very much like to see your schematics, once you're happy with them.  
Better late than never to learn about appropriate circuit designs.
----
John Dykstra                                jdykstra@nortelnetworks.com
Principal Software Architect                voice:  ESN 454-1604
Enterprise Solutions                        fax:    ESN 667-8549

Date: Fri, 14 May 1999 18:35:12
From: Dave Dameron 
Subject: Re: core... more

At 10:43 PM 5/6/99 -0400, Allison wrote:
> Well thanks to John D.  I have a bunch of cores from the early 60s.
>
> They are roughly 50mils x 11 mils x 15mils.  Cross section of the doughnut 
> is rectangular at 15x11 mils.  The reason for such rough measurements is my 
> vernier is only good to .001" and I'd need somthing fancier to be more 
> accurate.  By eyeball the 8e cores are smaller!  Based on several articles 
> I've read theses will need about 400->600ma to switch and will do so in 
> under 4uS.  I will have to test this in a jig.  Since the hole is ~20mils 
> several #40 wires should pass through it easily.

The previous mentioned article in Byte is "Coincident Current Ferrite Core
Memories" and is in the July, 1976 Issue. It is very helpful. It describes
the support electronics too (x,y drivers, sense, inhibit, and diode
steering logic) as the author had an intact core plane assembly, so did the
tests for switching currents, etc. on the intact assembly. The x,y drivers
used were TI SN75325 H-bridge drivers, on both ends of the x,y lines.
Alternate ways of running the sense lines are shown. Typically both ends
were terminated with 100 Ohm resistors and fed to a comparator with an
adjustable threshold, such as a SN7524.

-Dave

Date: Wed, 19 May 1999 10:48:26 +0200
From: wrm@ccii.co.za (Wouter de Waal)
Subject: Re: core... more

Hi Chuck and all

> Could somebody *please* scan this article in and post a URL to it? (I'll
> host it if it doesn't have web space (and I can scan it too if someone has
> the issue to loan or can make a good copy of it))

This just so happens to be one of the three BYTEs of that era that I
own. They're water damaged but readable.

http://ccii.dockside.co.za/~wrm/b
yte

And that goes over a 33K analog leased line shared by 20 people,
so be patient... :-)

Wouter

Date: Wed, 19 May 1999 09:50:43 -0700
From: Chuck McManis   | Block address
Subject: THANKS (and mirror site!) Re: core... more


THANKS Wouter! 

This is a wonderful article, I'm reading it now...

I've taken the liberty of mirroring these pages on my web server (T1 speed) at
        http://www.mcmanis.com/~
cmcmanis/core
 And put all of them together into a single zip file for easy downloading.

--Chuck

Date: Wed, 19 May 1999 09:53:35 -0700 (PDT)
From: Seth   | Block address
Subject: Re: core... more

On Mon, 17 May 1999, Seth wrote:

> On Mon, 17 May 1999, Chuck McManis wrote:
> 
> > At 06:35 PM 5/14/99 +0000, it was written:
> > > The previous mentioned article in Byte is "Coincident Current Ferrite
> > > Core Memories" and is in the July, 1976 Issue. It is very helpful.
> > 
> > Could somebody *please* scan this article in and post a URL to it? (I'll
> > host it if it doesn't have web space (and I can scan it too if someone has
> > the issue to loan or can make a good copy of it))
> > 
> > --Chuck

I've found my copy, and scanned it!

Check here:

   http://www.retronet.net/byte/


The files are fairly large, but you're warned what size they are
before you click on them.


Be gentle -- the server is a 486/33 running off my ADSL line :)  But
you should get at least 400kbps out of the line.

-Seth
--
"You know, if there's one thing more destructive      | Seth J. Morabito
 than a hydrogen bomb, it's a hydrogen bomb strapped  | sethm@loomcom.com
 to the back of a MANLY MAN!!!"  [James "Kibo" Parry] |   Perth ==> *

To Ethan's Home Page

HTML 2.0 Checked! Last modified: 29 June 1999
Compilation © Copyright 1999, Ethan Dicks <erd@infinet.com>. All Rights Reserved.