/LMAR=0/XLINE=3/FONT#0=BASL30/FONT#1=BASI30/FONT#2=BASB30/FONT#3=STA200/FONT#4=NGB25
␈↓ ↓H␈↓␈↓βS␈↓∧ Artificial Intelligence Laboratory, STANFORD UNIVERSITY, Stanford, California 94305
␈↓ ↓H␈↓∧Telephone 415 497-4971␈↓ 

August 5, 1975




␈↓ ↓H␈↓Mr. Art Hausman, President
␈↓ ↓H␈↓Ampex Corporation
␈↓ ↓H␈↓401 Broadway
␈↓ ↓H␈↓Redwood City, California 94063

␈↓ ↓H␈↓Dear Mr. Hausman:

␈↓ ↓H␈↓        We␈α∞wish␈α∞to␈α∞express␈α∞our␈α∞concern␈α∞about␈α∞what␈α∞appears␈α∞to␈α∞be␈α∞a␈α∞lack␈α∞of␈α∞quality␈α∞control␈α∞in␈α
a
␈↓ ↓H␈↓computer␈α
disk␈α
system␈α
that␈α
we␈α
recently␈α
ordered␈α
from␈α
your␈α
company.␈α
 We␈α
initiated␈α
a␈α
search␈α
for␈αa
␈↓ ↓H␈↓replacement␈α∞for␈α∞our␈α∂IBM␈α∞3330-11␈α∞system␈α∂several␈α∞months␈α∞ago␈α∞and␈α∂selected␈α∞your␈α∞product␈α∂on␈α∞the
␈↓ ↓H␈↓basis of price and ≡eld performance of the early installations.

␈↓ ↓H␈↓Based␈α∞on␈α∞your␈α∞quoted␈α∞delivery␈α∂date␈α∞of␈α∞June␈α∞20,␈α∞we␈α∂arranged␈α∞to␈α∞release␈α∞our␈α∞IBM␈α∂equipment␈α∞a
␈↓ ↓H␈↓month␈α∞and␈α
a␈α∞half␈α
later.␈α∞ This␈α
planned␈α∞overlap␈α∞has␈α
now␈α∞been␈α
eaten␈α∞up␈α
by␈α∞slightly␈α∞late␈α
delivery
␈↓ ↓H␈↓and␈αchronic␈α
malfunctions␈αin␈α
your␈αequipment.␈α We␈α
are␈αnow␈α
faced␈αwith␈αthe␈α
prospect␈αof␈α
having␈αto
␈↓ ↓H␈↓extend␈αthe␈αIBM␈αlease,␈αat␈αsubstantial␈αexpense,␈αand␈α
not␈αhaving␈αany␈αclear␈αidea␈αof␈αwhen␈αwe␈α
can␈αget
␈↓ ↓H␈↓out␈α∩of␈α⊃this␈α∩bind.␈α⊃ At␈α∩the␈α⊃present␈α∩time,␈α⊃the␈α∩control␈α⊃unit␈α∩is␈α⊃not␈α∩su≠ciently␈α⊃reliable␈α∩to␈α⊃sustain
␈↓ ↓H␈↓operation of our system.  We wish to call the following speci≡c facts to your attention.

␈↓ ↓H␈↓        1.␈α
 We␈α
have␈α
discovered␈α
several␈α
marginalities␈α
in␈αthe␈α
control␈α
unit,␈α
which,␈α
up␈α
until␈α
a␈αweek␈α
ago,
␈↓ ↓H␈↓caused␈α∞it␈α∞to␈α
be␈α∞totally␈α∞unusable.␈α
 In␈α∞fact,␈α∞one␈α
of␈α∞the␈α∞diagnostics␈α
(A4)␈α∞still␈α∞does␈α
not␈α∞run.␈α∞ It␈α
took
␈↓ ↓H␈↓Ampex␈α
≡eld␈α
engineers␈α
many␈α
hours␈α
of␈α
board␈αshu∞ing␈α
to␈α
≡nd␈α
a␈α
combination␈α
of␈α
seven␈α
cards␈αthat
␈↓ ↓H␈↓would␈α∂make␈α⊂the␈α∂control␈α∂unit␈α⊂run␈α∂at␈α⊂all.␈α∂ Despite␈α∂all␈α⊂e≥orts,␈α∂the␈α∂control␈α⊂unit␈α∂does␈α⊂not␈α∂perform
␈↓ ↓H␈↓reliably.

␈↓ ↓H␈↓        2.␈α Many␈αof␈αthe␈αcircuits␈αhave␈αthe␈αappearance␈αof␈αhaving␈αbeen␈αtranslated␈αfrom␈αECL␈αlogic␈αto
␈↓ ↓H␈↓TTL␈αwithout␈αadequate␈αconsideration␈αof␈αthe␈αe≥ect␈αof␈αchanging␈αthe␈αlogic␈αfamily.␈α Where␈αECL␈αhas
␈↓ ↓H␈↓complementary␈αoutputs,␈αTTL␈αrequires␈α
extra␈αinverters␈αand␈αhas␈α
slightly␈αlonger␈αgate␈αdelays,␈α
so␈αthat
␈↓ ↓H␈↓simple␈α∞substitution␈α∞of␈α∞TTL␈α∞circuits␈α∞causes␈α
critical␈α∞timing␈α∞chains␈α∞to␈α∞malfunction.␈α∞ Also,␈α∞we␈α
have
␈↓ ↓H␈↓informed␈α∩your␈α∩representatives␈α⊃that␈α∩we␈α∩believe␈α⊃that␈α∩the␈α∩cause␈α⊃of␈α∩one␈α∩of␈α⊃the␈α∩problems␈α∩is␈α⊃the
␈↓ ↓H␈↓incorrect copying of a signal name from the original IBM circuit.

␈↓ ↓H␈↓        3.␈α One␈αsymptom␈αof␈αshortcomings␈αin␈αthe␈αdesign␈αis␈αthe␈αnarrow␈αnoise␈αmargins:␈αour␈αcontroller
␈↓ ↓H␈↓malfunctions␈αconsistently␈α
when␈αthe␈αpower␈α
supply␈αvoltage␈αis␈α
adjusted␈αto␈αthe␈α
"normal"␈αTTL␈αlevel␈α
of
␈↓ ↓H␈↓5 volts, but works somewhat better when the voltage is lowered to 4.85.

␈↓ ↓H␈↓        4.␈α Read␈αampli≡er␈αcards␈αturn␈αout␈αto␈αbe␈αcritical␈αin␈αdetermining␈αerror␈αrates.␈α Some␈αcards␈αgive
␈↓ ↓H␈↓100␈αtimes␈αthe␈αerror␈αrate␈αof␈αothers.␈α These␈αrates␈αmove␈αfrom␈αdrive␈αto␈αdrive␈αas␈αthe␈αcards␈αare␈αmoved.
␈↓ ↓H␈↓Ten␈α
new␈α
cards␈α
supplied␈α
by␈α
the␈α
factory␈α
produced␈α
only␈α
one␈α
more␈α
good␈α
card␈α
with␈α
low␈α∞error␈α
rates.
␈↓ ↓H␈↓This␈α⊃means␈α⊂that␈α⊃production␈α⊃quality␈α⊂control␈α⊃is␈α⊃bad␈α⊂and␈α⊃that␈α⊃the␈α⊂factory␈α⊃tests␈α⊃are␈α⊂inadequate.
␈↓ ↓H␈↓Random␈α⊃≡xes␈α⊃like␈α⊃replacing␈α⊂ten␈α⊃percent␈α⊃resistors␈α⊃by␈α⊃one␈α⊂percent␈α⊃resistors␈α⊃were␈α⊃tried␈α⊃by␈α⊂your
␈↓ ↓H␈↓service␈α⊂man␈α⊂at␈α⊂factory␈α⊂suggestion␈α⊃and␈α⊂produced␈α⊂insu≠cient␈α⊂improvement␈α⊂suggesting␈α⊃that␈α⊂they
␈↓ ↓H␈↓αMr. Art Hausman␈↓ ¬hAugust 5, 1975␈↓ 
nPage 2␈↓ 

␈↓ ↓H␈↓don't␈α
yet␈α
understand␈α
the␈α
problem.␈α
 If␈α∞you␈α
are␈α
producing␈α
mostly␈α
inadequate␈α
read␈α∞ampli≡ers␈α
now,
␈↓ ↓H␈↓can't␈α∂test␈α∂them␈α∂and␈α∂don't␈α∂understand␈α∂the␈α∂problem,␈α∂then␈α∂you␈α∂are␈α∂risking␈α∂catastrophe␈α∂as␈α∂well␈α∂as
␈↓ ↓H␈↓incurring␈α∂costs␈α∂in␈α∂producing␈α∂unusable␈α∂cards.␈α∂ A␈α∂slight␈α∂change␈α∂in␈α∂your␈α∂production␈α∂process␈α∂may
␈↓ ↓H␈↓cause you to produce totally useless boards.

␈↓ ↓H␈↓        5.␈α One␈αof␈αthe␈α≡ve␈αdisk␈αdrives␈αthat␈αwas␈αdelivered␈αhere␈α(and␈αpresumably␈αhad␈αpassed␈αfactory
␈↓ ↓H␈↓inspection)␈α
was␈α
so␈α
mechanically␈α
unstable␈αthat␈α
doing␈α
a␈α
seek␈α
would␈α
cause␈αit␈α
to␈α
be␈α
unable␈α
to␈αread␈α
any
␈↓ ↓H␈↓data␈α∞previously␈α∞written␈α∞and␈α∞to␈α∞write␈α∞≡les␈α∞that␈α∞could␈α∞not␈α∞be␈α∞read␈α∞later.␈α∞ It␈α∞was␈α∞removed␈α∂and,␈α∞to
␈↓ ↓H␈↓date, has not been replaced.

␈↓ ↓H␈↓        We␈α∂were,␈α∂and␈α∂are,␈α∂attracted␈α∂by␈α∂your␈α∞lower␈α∂price␈α∂than␈α∂IBM's,␈α∂but␈α∂we␈α∂are␈α∂quite␈α∞unhappy
␈↓ ↓H␈↓about␈αthe␈αamount␈αof␈αtime␈αour␈αengineers␈αhave␈αhad␈αto␈αspend␈αtracking␈αdown␈αyour␈αbugs.␈α We␈αhope,
␈↓ ↓H␈↓Mr. Hausman, that you will give this matter careful attention.


␈↓ ↓H␈↓Sincerely,



␈↓ ↓H␈↓John McCarthy
␈↓ ↓H␈↓Director
␈↓ ↓H␈↓Professor of Computer Science







␈↓ ↓H␈↓cc:  Hersche Allen, Les Earnest, Ralph Gorin, Ted Panofsky, Je≥ Rubin,
␈↓ ↓H␈↓        Rainer Schulz (IMSSS), Richard High≡eld (Purchasing)

␈↓ ↓H␈↓Stanley Mantell
␈↓ ↓H␈↓Vice President and General Manager
␈↓ ↓H␈↓Data Products Division
␈↓ ↓H␈↓Ampex Corporation
␈↓ ↓H␈↓Redwood City, California 94063

␈↓ ↓H␈↓Robert M. Trick
␈↓ ↓H␈↓Northwest Regional Manager
␈↓ ↓H␈↓Data Products Division
␈↓ ↓H␈↓Ampex Corportion
␈↓ ↓H␈↓465 Maude Avenue
␈↓ ↓H␈↓Sunnyvale, Calif. 94086