IBM 1620

[Tothwolf]

On Sun, 30 Aug 2015, Chuck Guzis wrote:
> On 08/30/2015 11:15 AM, Paul Koning wrote:
>
>> So there'd be what, 120,000 cores to thread?  That might be a bit 
>> daunting from a human-hour standpoint.  I'll wager that 120K cores 
>> wasn't even a day's output for outfits like Fabritek.
>> 
>> Those cores weren't threaded one by one.  You'd start by setting the 
>> cores into a holding jig, which positions them correctly.  Then you 
>> thread wire from edge to edge.  The article mentions a needle with the 
>> wire welded to its end; that makes sense because the copper wire is 
>> unlikely to be stiff enough.  So the number of individual threading 
>> operations is 3-4x the square root of the core plane size. For example, 
>> on a 4k core plane, it would be 200 steps, give or take. (A bit more on 
>> a CDC 6000 series core plane with its peculiar 5 wire architecture.)
>
> I've seen the archival photos.  Still, the possibility of missing or 
> damaging a single core was always there.  I wonder what the rejection 
> rate was.
>
> I have heard of machine-fabricated core as well, but I thought that it 
> only applied to larger, slower bulk core store.  One can certainly 
> understand why plated-wire or thin-film technologies were attractive.
>
> Sigh.  Another lost manual art.  I can remember during the 70s that the 
> hot thing was to learn IC layout--the wives of a number of co-workers 
> were going to night courses for that.

According to IBM's paper, extra cores were threaded initially and used to 
replace any cores that tested bad during assembly. Any extras were crushed 
after assembly to remove them.