Path: utzoo!attcan!uunet!lll-winken!lll-tis!helios.ee.lbl.gov!pasteur!ucbvax!hplabs!hpda!hpcuhb!hpihoah!jwatts
From: jwatts@hpihoah.HP.COM (Jon Watts)
Newsgroups: comp.misc
Subject: Re: Core memory
Message-ID: <4620001@hpihoah.HP.COM>
Date: 20 Jul 88 01:42:40 GMT
References: <1486@crete.cs.glasgow.ac.uk>
Organization: Hewlett Packard, Cupertino
Lines: 63

Up until about 2 months ago I worked for a company that makes (note *present*
tense :-( ) core memories.  I worked on core memories extensively while there.
     Most core memories used by the computer industry were hand strung but the
newer ones are machine strung (machine stringing was introduced around 1978).
In a hand strung memory the cores are loaded by operators using tweezers and
microscopes and threaded with tiny needles.  In machine strung memories the
cores are shaken into a fine grate which aligns them in the proper orientation.
In the ones my company made (I don't know about any one elses) one edge of the
each core was allowed to sink into a sheet of unvulcanized rubber which was
then cured to cement the cores in place.  The core mats then were strung by a
machine designed specifically for that purpose.  The way it works is the
machine first welds needles to the ends of the select wires.  The needles are
long enough to pass all the way through the core mat, come out the other side,
and be grabed by the machine again.  The needles are held in a pair of jaws
which has a groove for each needle.  The jaws push the lines through a core
mat until it is grabed on the otherside by another set of jaws this is repeated
until the wires have passed through all the core mats in the assembly
(typically 1 mat for each bit in the word width). After all the mats have been
strung along one axis the process is repeated for the other axis and then for
the sense lines and inhibit lines (if present - some designs use the same lines
for both sense and inhibit) after the lines have been strung they are
terminated by soldering them to pads on the board which connect them to the
drive and sense electronics.  The soldering is semi-automatic, the operator
must align the soldering head but the head heats pulses to remove insulation
and solder following a programmed profile.  Once assembled the core assembly
is acordian folded to make a more compact package.
     The sizes of the components involed are truly amazing the newer of the
designs I worked on used 13 mil cores (that's 0.013 inch outside diameter - 
approximately 7 mils I.D.), was strung with 44 and 48 gauge wire and contained
aprox. 600,000 cores (32K x 17 bits plus spare rows and columns).  The
assemblies I worked with were for a military application and so operated from
-55c to +71c they cost aprox. $15K to build and sold for aprox. $25K.  The
physical size is aprox. 6"x4"x2.5" plus and external card aprox. 6"x4" is
needed for every two memory assemblies.
     I believe the statement that the shuttles main computers use core is
correct.  They also use relatively old core technology.  I heard they are have
4Kwords of memory (16 bits plus EDC I think).  Cores have several advantages
that the military and aerospace industries like and semiconductor memory
technologies are only in the last 2 or 3 years beginning to suppant them.
First of all they are non-volatile, you can leave a core memory with data in it
sit for years and it will still have the data when you come back.  Secondly
they are very resistant to external influences (read rad-hard), as long as
there is no current in the drive and sense lines you won't switch a core; and
the current needed to switch is on the order of half an amp far, more than most
noise.  They are read/write with unlimited write cycles and read and write take
the same amount of time (750 nS in the example above).  They also have good
reliability (amazing I know, I don't understand it but if you get one working
in the first place they just won't die).
     On the down side the cost is very high (see above).  They drink power
(the example system is about 15 or 20 watts).  They are large and heavy.  They
are slow.  And they are extremely difficult to design.  
     I am very pleased to say that the military has finally come to their
senses and started to allow the use of battery backed CMOS static RAM in place
of cores.  It took a long time because the military does not like batteries
(they create logistics problems since they need periodic replacement and are
potentially hazardous), but they are now allowing cores to pass into the
history of computers rather than remain a living dinosaur.

+-------------------------------------------+
|These opinions do not neccessarily reflect |
|those of my employer, my country or anyone |        -Jon Watts
|else if fact they aren't even opinion in   |
|fact they don't even exist, neither do I   |
|your hallucinating again. Is this Kansas?  |
+-------------------------------------------+