Path: utzoo!attcan!uunet!vsi!friedl
From: friedl@vsi.COM (Stephen J. Friedl)
Newsgroups: comp.arch
Subject: Re: Bubbles (was Re: Solid State Secondary Storage)
Summary: A possibly-correct lesson in bubble memories
Message-ID: <1012@vsi.COM>
Date: 14 Jan 89 22:16:10 GMT
References: <248@vlsi.ll.mit.edu> <2409@garth.UUCP>
Distribution: comp
Organization: V-Systems, Inc. -- Santa Ana, CA
Lines: 85

In article <2409@garth.UUCP>, phipps@garth.UUCP (Clay Phipps) writes:
>
> Bubble memory is nonvolatile. [...]
> 
> The major concern, as I recall, was that in the late 1970s,
> the error rate for bubbles was decimal orders of magnitude 
> worse than disks.  And disks kept getting better and better.
> 
> What ever happened to bubble memory, anyhow ?

[warning: vague memory alert here.  It's been a while...]

I worked on a bubble memory system back when I was at Telxon and
Intel was in the bubble memory business.  Their controller was
not so hard to use, but remember that bubbles are not random-
access.  Bubble memories are organized as lots of circular
storage loops, and an electromagnet embedded in the bubble unit
shifts all the loops around in step with a clock.

            _____ your data here
          _|_
         /   \   _________________________________
         |   |__/                                 \
         |   |  \_________________________________/
         |   |   _________________________________
         |   |__/                                 \
         |   |  \______________________A__________/
         |   |
         ~   ~   _________________________________
         |   |__/                                 \
         |   |  \_________________________________/
         \___/

        transfer           storage loops
          loop

(I don't know if these are the right terms, but I'll use them
assuming they are).  For example, to get the data marked "A"
above, you shift all the loops around until A is at the transfer
point, and then move the data to the transfer loop.  Then rotate
the transfer loop around until your data is at the output point, a
bit at a time (the controller turned it into a byte stream).

There were lots and lots of loops -- hundreds or thousands -- and
they all had synchronization patterns embedded so you could find
the "start" of the loop in case you powered down in the middle of
a rotation.  These are easiest to use if you pick some pagesize
that is probably related to the number of storage loops, so that
each bit in your page goes in a different loop, but I'm really
unsure about this.

The bubble memories themselves were about an inch and a half
square, maybe a quarter inch thick, and they had a heavy case
that held the magnet.  The controller took a lot of power (I
think it was NMOS), as did the bipolar drivers.

I believe that bubble memories are inherently rad-hard, but not
necessarily the controller chips.  I really don't believe,
though, that they were "decimal orders of magnitude" less
reliable than disks (Clay *did* say late 70's, and my
use was in 1981).  If anything, I would guess it's the other
way around.

Texas Instruments and Hitachi (and maybe others) had bubble
chipsets, but I know nothing about them.  Intel spun off/sold
their bubble unit to another company, I think something like
Magnesys, and I think they are in Sacramento.

We never went ahead with the project because CMOS static RAM was
coming down in price, and the power requirements for the bubbles
was just to onerous for a handheld unit. Too bad, I was really
getting into my little 8051 design lab :-(.  After that, I started
writing compilers and it marked the end of my hardware career --
double :-(.

I'm really digging deep to come up with this info.  Anybody out
there remember better than me?  Flames are welcome if you must.

     Steve

-- 
Stephen J. Friedl        3B2-kind-of-guy            friedl@vsi.com
V-Systems, Inc.        I speak for me only      attmail!vsi!friedl
Santa Ana, CA  USA       +1 714 545 6442    {backbones}!vsi!friedl
---------Nancy Reagan on Hawaiian musicians: "Just say Ho"--------