: :
e:
rote:
ux Vbias
a mosfet.
ds with their fake dropout specs while conveniently omitting the fact that Vbias must be greater than Vout + 1.5V.
e millivolts
r an opamp
down to
showy 80dB PSRR at 100 Hz. Battery operation usually doesn't care a whole lot about PSRR. And the thermal impedance specs are so bad, you just try ge tting 800mA out of it with any kind voltage headroom without using a liquid nitrogen drip.
es 800 mA
at much
x-1
25mV 12Vhe
ve used
mostly.
le
-channel
ilinx
g a few
ve but
elves all
noxious and
:(
, and
brave at
r,
FBI on
of times
on) as
guess he was close to not being with us any more
we like it
took over at that point
simulate the design, and then had another designer lay out the chip
ior are
crete component designs.
high frequencies, the fact that the connections are lot shorter (and less inductive) makes a big difference. Bob Widlar and Barry Gilbert were famous ly good at exploiting the advantages these sorts of difference offer.
you
re
ance
tly
e
's a
ad
ed
ps
oard
in
e to
d.
they
eFilotex coaxial cable (about 1.1 mm OD IIRR). The other thing that messed u p the board was Gigabit Logic's failure to put the input capacitance of the ir logic on the data sheet - on rev b every last clock input had it's own p rivate driver.
ong 100k ECL synchronous counter at 200MHz, which would have been doable wi th careful design - as I had spelled out in the specification. The guy who did the detailed design of the board was entirely capable of managing that, but was also smart enough to know that the machine was never going to get into production - it didn't - and didn't bother.
ted circuit layout I might well have caught it, but our idiot project manag er didn't want to slow down the process by the day that would have taken.
k much longer to find (but was much easier to cope with when it did get not iced).
three years work when the machine was working - to take away three years w orth of weekly reports. About fifteen years later I got around to scanning them.
Semi-manual. We never got enough boards made that the people who loaded the m would have automated the placement. I got fairly close to that - these we re the first surface mount boards used a Cambridge Instruments - and Iened up being responsible for getting the tools we needed - which was mainly a G roatmoor hot air reworking tool, which I'd seen at every service mount shop that I'd visited.
sNo. We bought our supplies off the shelf, and they were perfectly conventio nal switching power supplies, if reasonably quiet.
No idea.
eDelivering information to a human observer can take advantage of the relati ve slowness of the human comprehension system where it's taking up complica ted information.
l
When we wanted speed we went for ECL and GaAs. Most of the rest of the mach ine was upgraded - as little and as cheaply as possible - from original Lin tech electron beam tester. We did go for the AMD Taxichip links for high vo lume data transfer - and put minimal galvanic isolation into the links to m inimise ground loop problems.
t
f n
s
It was an up-market board house, and they didn't have any trouble with the layout design rules. In fact they probably prescribed them.
Printed circuit layout was done by specialist draftsmen and some of our was done by a specialist sub-contractor, but the engineers got to sit with the draft's person to sort out the more demanding bits of the boards. Given th e chance, I would have reviewed every last board layout before we sent it o ff to get etched, but our idiot project manager didn't want to delay the pr ocess by delaying sending it out for that.
d
Quite a lot easier. Most of the logic could have been squeezed into program mable logic parts. I did a detailed design on a similar sort of system at N ijmegen University around 1997, and the programmable parts we could buy the n still weren't fast enough for the fastest bits but did allow us to squeez ed down the data handling area quite a lot.
We used a lot of 22V10 programmable parts in 1989, and even then we could h ave done better, but it would have meant chucking out existing designs.
The ICT Place ICT7024 was a drop-in replacement for the 22V10 but a good de al more powerful, but I didn't get use one until 1991 when I was wording fo r another company.
s
The AMD Taxichip was a long step along the road to USB, and we were really very happy to have something that could run a serial link at 125MHz. USB-C would have put us in the seventh heaven.
Motorola were starting to push ECLinPS back in 1990 (and one of their chips does get mentioned in one of my later weekly reports) but you couldn't buy them then.
By the time I'd got to Niimegen - at the end of 1993 - ECLinPS was stocked by major distributors , and I used some of the parts to upgrade an old elec tron spin resonance machine, and was planning to use quite a few more in th e new version that got designed but never built.