MODERN REPLACEMENTS FOR IBM 091 TRANSISTOR
AS USED IN THE 1620 DATA PROCESSING SYSTEM

SUMMARY AT THE END

Updated August 2021 for parts availability

Used in the Type FM Current Driver card.
Also applies to the DKA card in the 1401, which is effectively identical.
Also applies to the DFV/DFW/DFY cards in the 1410.

The 1410 needs one of the heat-sinked approaches,
and the DFV card requires 100V.  I suggest 2SC5706 for all.

I have seen a photograph of an FM card with a Type 087 transistor;
therefore, my research below probably applies to 087 as well as 091.

March 2015
May-August 2016
2019

May 2017 -----------------

My 1620 core array has failed!  Two open-circuit Y half-select
lines, 0xx52 and 1xx98.  I will try to repair them sometime.
Meanwhile, I'm using the solid-state ram board.

It's easy to confirm on the scope, just probe across the
suspected matrix switch output winding, with the 1620 set to
hammer on the bad address.  Good = 5V pulse, bad = 25V pulse.
Ghod knows what it's doing at the transistor.

TODO:

1. Disable matrix switches.
   Ground pin C of at least one MSA and one MSD current driver.

2. Repair core.  I don't want all my transistor work to become
irrelevant.  There's no sign of corrosion; I think it just broke
from 60 years of thermal cycling.  Probably right at the terminal.
Even so, disconnecting the stack from 200+ wires will be tedious.
Maybe I can get by without disconnecting.
This will be a good winter project.

3. Create as-built ram board schematic.
--------------------------

I thought these would also work for 086 (FL/ARW Decode Switch) and
071 (WJ MARS/Inhibit Driver), but most parts suited for FM are
too slow for FL, and those suited for FL are too slow for WJ.
See Decode_and_Inhibit_Candidates.txt .  There is no one sweet spot
for all, it mostly takes a different type for each job.

Note: Silicon in FL requires a 9V zener in the emitter lead to prevent
B-E breakdown.  The zener must be an ordinary diode drop in the forward
direction.  It must also "reverse recover" faster than the B-E diode.

All three cards switch 300mA, but 086 switches "cold", and
071 has a lower Vce (12V excluding transients).  Therefore,
086 and 071 max ratings are not as stringent.  But they are
saturated, and need to be fast even so.  Especially the 071,
because Ibb2 is low.  That means they have to be more agile
and therefore smaller.

CAVEAT!  Some of these substitutions have not been tested yet.

The Current Driver card has more heat sink than Decode and Inhibit.
Due to 091's unique heat spreader design, the IBM heat sink is
incompatible with all substitutes.  But power dissipation is less than
one watt, and 091 needs the big sink only because it's Germanium.
Silicon parts don't need much, especially with forced-air cooling.
I suggest some models below anyway.

Beta = 30 is the theoretical minimum.  I insist on 40, at 1V/300mA.
The part I am currently using is the TO-220 D44C8, on a small heat sink
modified for low profile.

The FM card was designed to get 1V when beta is 100.  The limiting
case is 1.6V and beta 30.  With Si, this shifts to 1.4V and 1.8V.
(1.1V when beta is infinite.)
If you decrease the 91-ohm resistor to 68 (e.g. by paralleling 270),
Vce shifts back to the design values of 1V and 1.6V .

The bad 091 measures 35pF at 9V.
D44C8 Cob is 100pF max at 10V; measures 45pF at 9V.  Anything less should be okay.
D44H11 Cob measures 65pF at 9V, so more is going on than just Cob.
D44C8 fT is 40MHz max at 20mA/4V.  We want the spec at 2V/300mA.

NOTE: Since the circuit reverse-biases the B-E junction, BVceo is not relevant.

The voltage specs below are BVcbo.  I assume no punch-through.
(If BVces == BVcbo, that's guaranteed.)

Distinguishing Avalanche Breakdown and Punch-through

Avalanche won't oscillate or pull to ground without shunt capacitance.
Punch-through falls right to ground, super fast (1ns),
and oscillates without capacitance.
My 2N1304 avalanches (around 75V), my BC337 punches through (around 150V).
I read a patent for adding controlled punch-through to a process,
that said it gave better reliability than avalanche.  Motorola?
Can't remember.  Anyway, maybe all modern parts are built to punch through.
I used a HV power supply and 10K 10W to collector, base+emitter grounded,
and 15K 1W from collector to scope with 50-ohm termination.  Transistor
gets hot, run for one minute unless cooled.

WARNING - If you use a sampler, include at least 20dB of 50-ohm attenuation
ahead of the series resistor, to prevent the transient coupled through
the resistor's parasitic capacitance from blowing the sampling diodes.

Look at this again at DC.  Punch-through is happening when emitter and collector
rise and fall together, one volt change at C causes one volt change at E.
Compare to Beale et. al. who say that punch-through is when reverse-biasing
the base doesn't stop the oscillation. 

JUST LEAVE IN THE AIR

Absolute minimum tooling, labor, and materials.
These parts have beta to spare and will run less than 1.4V / 0.42W / 120C Tj.
And that's in still air at 35C 95F i.e. fan failure.

 TO-92 SC-71 E-Line

BC337      50V 0.8A SOA .2A@45V .3A@40V
KSD1616A  120V 1A   SOA .15A@60V .3A@40V  Works in FM card
MPS650     60V 2A
MPS651     80V 2A   SOA .4A@60V Works in FM card
STSA1805  150V 5A   Slightly slower than D44C8, should be fine
ZTX449     50V 1A
ZTX650     60V 2A
ZTX651     80V 2A   SOA 1A@60V
ZTX692B    70V 1A   SOA .7A@70V
ZTX851    150V 5A   SOA 1A@60V
ZTX1053A  150V 3A   SOA .6A@75V
ZXTN2010A 150V 4.5A SOA 1A@60V

NEAR ORIGINAL LOOK

Pot the above in the original heat sink.
It's a lot of work, and I don't know how much heat it can dissipate.

  PROCEDURE 1

Make a 0.100" thick heat spreader disk that's a smooth tight fit in the heat sink.
3/8" rod; grind down to 0.325".
Cut a radial slot for the transistor leads.
Cover the bottom with tape, slit to match the slot.
Wrap the transistor leads for 0.125" with a strip of tape.
Wrap a tube of tape around the disk.
Stick the transistor in the slot.
Fill the annular space around the transistor with thermal epoxy.
After it sets, peel off the tube of tape.
Form the transistor leads to hit the TO-5 pads.
Coat the heat sink cavity lightly with thermal grease.
Observing lead orientation, insert the potted transistor into the heat sink.

  PROCEDURE 2 (more steps but no machining)

Find a straw slightly larger than the hole.
Snip off 3/16".
Coat the hole with thermal grease, then wipe off gently leaving a thin film.
Cover the hole on the bottom with Scotch Magic Tape et. al.
Install the heat sink.
With a needle or pin, poke small holes for the transistor leads.
Form the transistor leads to match the old TO-5 EBC pin circle.
Slip the transistor into place through the holes but do not solder.
Seal the holes with dabs of white glue on the end of a toothpick.
Coat the inside of the straw piece with thermal grease.
Put a bead of white glue on the bottom edge of the straw piece.
Position the straw piece on the heat sink and secure with tape.
Wait for the glue to dry.
Remove the tape securing the straw piece.
Fill the cavity with epoxy.
Wait for the epoxy to set.
Solder the transistor leads.

SOLDER TO COPPER-CLAD PCB (through hole leads)

TO-251

Drill mounting holes and one big hole, attach wires
and cover with tubing, solder wires into SMS card.

2SC4135  100V 2A 15pF typ
2SC5706  100V 5A SOA 1A@100V 15pF typ

SOLDER TO COPPER-CLAD PCB (surface mount leads) (no testing done)

TO-252

2SC5706
2SD1816  120V 4A SOA .5A@50V 40pF typ
2SD2318  NRND
MJD44H11  80V 8A 45pF typ, not as slow as D44H11
NSS1C301ET4G
STD1802  fT at 50mA

SOT-223

2STN1360
DNLS350E
FZT651
NSS1C201MZ4 (or NSV)
NZT560
NZT651   (Fairchild tech support says fT is 50MHz min at 500mA)
NZT902
NZT6717
PBSS4350Z
PBSS8110Z
PZT651
STN851A

METAL CAN

TO-39

Minimal tools and labor, but parts are expensive.
The parts below are untested.  If they are too fast,
you can add a slowdown cap on the component side,
provided the holes are big enough for two leads.  Mine are.
No heat sink; if TO-92 doesn't need it, we don't either.

 NTE16005 claimed sub but might be slow like 2N3421
2N5320   100V 2A CEN $3.09 Beta reads 65 at 300mA/1V, maintained down to 0.3V
2N5321    75V
2N3507      MOT see Motorola-SeminarsandApplicationBooksSmall_Signal_Transistor_Data-DL.pdf $$$
2N3735      MOT Motorola-SeminarsandApplicationBooksSmall_Signal_Transistor_Data-DL $$$
2N4895      $$$

BOLT TO COPPER-CLAD PCB OR LOW-PROFILE HEAT SINK

Wakefield 242-125ABE-22 or 274-3AB,
Aavid 592502B03400G or 577002B00000G,
CUI HSS-B20-0635H,
cut off mounting tabs, and elevate with nut plus washer to avoid
scratching the SMS card.  0.125" brings total height to 0.346",
which is less than Type 028 (tall TO-5) height of 0.385".
60K rise at 2W natural convection.

TO-126/TO-225/SOT32

BD137     60V
BD139     80V 1.5A SOA 1A@80V
MJE180    60V
MJE181    80V
MJE182   100V 3A   SOA .2A@80V .3A@60V .8A@40V Works in FM card

TO-202

CEN-U05   60V 2A
CEN-U06   80V
CEN-U07  100V

TO-220

D44C8     60V 4A 100pF max   In use.  Speed equals 091.  NOTE: EOL!

DISQUALIFIED

 QUASI-SATURATION

BC639         1A   (observed on curve tracer)
KSC1008       0.7A (observed on curve tracer)
MJE243        4A   (observed on curve tracer)
PN3569        0.5A
2SC3902       1.5A
2SC4027       1.5A
2SC4134
2SC4135
2SC4488  120V 1A
2SC4614       1.5A
2SC5171       2A
2SC6097   60V 3A
2SC6098
2SC6099
2SD1733
2SD1815  120V 3A

 INSUFFICIENT BETA

(Some of these parts may work in FL or WJ; FM provides less bias.)

2N1613
2N1711
2N2102 
2N2219A
2N3053
2N4401 
2N5681/2
2SC3332
BD241A/B/C
KSC2331     Works in 1620 - barely
ZTX450
ZTX451
ZTX452
ZTX453
ZTX454
ZTX455

 SLOW (fT or Cob)

2N3420   150pF
2N3421   125V 2A 40@1A/2V 40MHz 150pF CEN $3.33 19V on test jig 2
2SC6082
2SC6144
2STN1550 fT not given
BC141    Tested slow
BCP55/6  NXP: fT at 50mA; ON: steep drop after 100mA
BD237
BD239
BD787    4A. Jig 2: Ts=200ns Tf=200ns 20V
D44H11   Works in 1620!  But slow.
KSP05/6  Steep drop
MPSA05/6 Steep drop
KSC2316  Tested slow
KSC2334  fT/Cob not given
KSC2383  Tested slow
KSC2690  Tested slow
KSD526
KSD1588
KSD1691  fT/Cob not given
TIP29A/C 3MHz

 INSUFFICIENT POWER DISSIPATION

2N1304   150mW

 INSUFFICIENT BVcbo

Since FM reverse-biases by one diode drop, the

criterion is somewhere between BVces and BVcbo.

The emitter is at -12V, so a 30V spike is 42V.

I think 60V is fine, maybe 50V.

MPSW01    40V

 DISCONTINUED

2N3725    EOL 80V 1.2A CEN $4.24 EOL 2021-03
2N4896
2N6715
2N6716    2A Use MPS651
2N6717    80V 1A Use MPS651. 1800 in stock, $1.62 . 
2N6718    2A Use MPS651
2SC2655   50V 2A
2SC4487   60V 3A
2SC6017  EOL 100V 10A SOA 3A@50V 60pF typ, might be slow
2SC6043   80V 2A
2SD1207   60V 2A
2SD1683
2SD1801  EOL 60V 2A 12pF typ
2SD1802  EOL 60V 3A 25pF typ
2SD1803  EOL 60V 5A SOA 2A@50V 40pF typ
2SD1816  OBSOLETE 120V 4A SOA .5A@50V 40pF typ
2SD1835   60V 2A
D44C8     ON discontinued in 2018, NTE still has D44C11 in stock but do not use NTE377 cross, it's D44H11
MPSW05
MPSW06
TN2219A
TN6715A

Since the Current Driver card is unsaturated,
the saturated switching speed is irrelevant.
But fT (at 2V/300mA) and Cob are important.
And there must be no quasi-saturation.

>>>> 086 (FL) and 071 (WJ) are used in saturated mode!

Claimed to be ok by Bob Hunter replacement list:

071 2N377 2N647 2N1302
086 2N214 2N557 2N1302 2N1304
091 2N214 2N228 2N557 2N1302

I have no direct experience with the following IBM parts
but believe the suggested replacement is not adequate.

092 2N228 but IBM-StandardModularSystem-Neff7.pdf says 1250mA (pg 141)
093 2N228 but IBM-StandardModularSystem-Neff7.pdf says 400mA (pg 220)

From Towers International Transistor Selector:

2N214  TO-22 40V 100mA 44min@35mA 300kHz
2N228  TO-22 40V 100mA 80tp@1mA   200kHz
2N377  TO-5  25V 200mA 20-60@30mA 3MHz (2N377A is 40V)
2N557  TO-5  20V 200mA 20mn@1mA   fT not specified
2N647  TO-1  25V 50mA  70tp@50mA  fT not specified
2N1302 TO-5  25V 300mA 20-80@10mA 1MHz.
2N1304 same except 40-200 and 4MHz

From New Jersey Semi datasheet

2N1304 300mA absolute maximum.  Beta@200mA is 15 minimum 110 typical.

071, 086, 091 require 300mA so 2N214 2N228 2N557 2N647 disqualified
086 sees 30V transients so 2N130x disqualified
091 sees 42V transients so 2N130x disqualified

As you can see, the Hunter list is wrong except possibly 2N130x for 071.
No, it's wrong there too: in datasheets from Central Semi, New Jersey Semi,
and RCA, beta is 15 or 20 at 200mA so it will be less at 300.

Unless you select for BV, hFE, and speed.  My West German-made 2N1304's
are fine for FL.  (But too slow for FM and WJ.)

TO-5 (except as noted) NPN Germanium with Vcb >= 30V and Ic >= 300mA

2N356A 30V 500mA 20-50@100mA 1MHz
2N357A 30V 500mA 25-75@200mA 2MHz
2N358A 30V 500mA 25-75@300mA 4MHz
2N440A 30V 300mA 40mn@50mA   5MHz
2N576A 40V 400mA 20-60@400mA 3MHz >>>>MAYBE
2N625  40V 500mA 20mn@500mA  -    TO-8
2N821  30V 400mA 40mn@50mA   -    OBS PKG
2N822  30V 400mA 40mn@50mA   -    OBS PKG
2N1170 40V 400mA 20mn@200mA  3MHz

None of these has enough BV to replace 091, but
2N358A or 2N576A can probably replace 071 if fast enough.

Silicon from Towers:
(All disqualified.)

2N697   60V  500mA 40-120@150mA 50MHz Insufficient beta at 300mA
2N1613  75V  600mA 40-120@150mA 80MHz
2N1711  75V  600mA 100mn@150mA  70MHz
2N2102  120V   1A  40-120@150mA 60MHz (Typo, Towers says 10mA)
2N2219A 75V  800mA 100mn@150mA  300MHz
2N3053  100V 700mA 50mn@150mA   100MHz

Germanium from D.A.T.A. 1962

Line Part
885  2N1170
1586 2N1473    40V 400mA 50@400/0.6 15pF 8Mc(f alpha b)
2293 2N1965    Error - this is silicon, and not TO-5

CLAMPING ANALYSIS

The FM card puts 12V through 510 ohms, then clamp diode, then 91 ohms,
then 8mA sink to -36V.  If you assume .3V diode drops and beta=100,
it balances at Ib = 300/100 = 3mA, so 11mA through 91 ohms.
.3+(.011*91) = 1.3V .  Assume 0.7V drop in the driver.
(12-1.3-.7) = 10V drop across the 510 for 19.6mA, which
implies 8.6mA through the clamp diode, whose .3V drop means
Vc = 1V .

Repeat for Silicon.
.7+(.011*91) = 1.7V .  Assume 0.7V drop in the driver.
(12-1.7-.7) = 9.6V drop across the 510 for 18.8ma, which
implies 7.8mA through the clamp diode, whose .3V drop means
Vc = 1.4V .

Clamping failure occurs when the diode conducts negligible current.
I91 = (12-.7-.3)/(510+91) = 11/600 = 18.3mA.  Ib = 18.3-8 = 10.3mA .
.3V+(.0183*91) = 2V, which .2V diode drop reduces to Vc = 1.8V .

At clamping failure, beta is 300/10.3 ~= 30.

Repeat for Silicon.
I91 = (12-.7-.7)/510+91) = 10.6/600 = 17.7mA or Ib = 9.7mA.
.7V+(.0177*91) = 2.3V, which .2V diode drop reduces to Vc = 2.1V.
Beta is 300/9.7 ~= 30.

TEST JIG 1

2N4401 driver.  Emitter at -3V.  Base drive is +/-5V through 1K.
180 ohm collector load to +5V.  One Si diode to DUT base then 470 to -3V.
DUT collector has Si diode to driver collector, and 51 ohm 2W carbon
comp load to +17V.  10uS active-low pulse (DUT usually off)
and 20kHz rep rate to avoid overheating Rload.

Ib(supply) = (5V-1.4)/180 ohms = 20mA.  Ibr = 3.7V/470 ohms = 7.9mA ~= 8mA.
If beta = 100, Ib = 3mA, Id1 = 11mA, Id2 = 9mA, and Vc = .7V.
At failure, Ib = 20-8 = 12mA, Id1 = 20mA, Id2 = 0, Vc = 1.4-.5 = .9V ~= 1V.
Looks like this is off a bit, because everything reads .8-1.1V with 1.1V
still showing clamp action.

READINGS

Note: The 091 data below is an old measurement on an FM card, so
it is not exactly commensurable.  Tf is a guess based on voltage
spike height, which equals D44C8.  Ts is fudged to follow the
difference between other Ts readings here and their analogues
in the old FM test run.

Any part with short Tf can be slowed down with a Miller cap.
For example, 36pF on BD139 gives 70ns.

Sorted by Tf

Type       Ts   Tf   Vc

091       130   90   1.0
086        80   230   .7

BD139      30   20   1.0
MPSW3725   45   30   1.0
MPS651     45   40    .9
TN2219A    75   40   1.1
TN6715A    50   45    .9
2N2219A    75   45   1.1
PN2222A    90   50   1.0 Slow start
KSC2331   120   50    .9
2N4401    100   55   1.0 Slow start
KSD1616A   40   60    .9
D44C8      70   90    .9
------------------------
STSA1805   60  120    .8
MJD44H11   80  140    .9
TN3019A   250  180   1.0 Slow start
BC141-16  200  250   1.1 SLOW start
D44H11-ST 120  300    .8 Works in 1620 but don't know how well
2N1304     80  320   1.1


REAL FM CARD

With Milwaukee resistor pullup instead of current source.
The numbers come from scoped voltage rise at the load resistor.
The speedup cap saturates the transistor for about 400ns.

Type       Ts   Tf

091-1     100  120       Cards from K2 and H3
091-2     100  200       Card from K28

2N5320      0  100
2SD1803     0  140
STSA1805    0  160
MJD44H11    0  160
D44C8-GE    0  180       Card from K27
D44H11      0  400
2N3421      0  450

TEST JIG 2

This is closer to real life.
2N4401 driver.  (15 ohms in collector, does not affect FL or WJ mode.)

Emitter at -3V.  Base drive is +3V/-7V through 270 ohms.
160 ohm collector load.  Two Si diodes to DUT base and 470 to -3V,

to check temperature, or one diode to check clamping.
DUT collector has Si clamp diode to driver collector, and 56 ohm 20W WW,

Milwaukee type 43X302 load resistor to +17V.
+4V (one diode) or +4.7V (two diodes), -3V, +12mA max, -8mA.
50kHz rep rate, short active-high pulse (DUT usually on to maximize

saturation), adjust Vcc for 300mA average current,
adjust pulse width to get one peak plus a bit.

This is slower than real life: Tf for D44C8 is 120ns and Vspike is
only 30V across the transistor whereas real Vspike is 30V across
the load and 42V across the transistor.  Maybe the resistor's Q is
lower than the 1620.

To avoid extra stress on the Decode Switches, on fast parts
we add a Miller capacitor to reduce the spike to D44C8 size.
(Which in turn matches IBM 091.)  I use this jig to pick the cap.

MPSW3725   47pF
2N2219A    33pF (45C rise with heat sink)
BC337      33pF
MPS651     33pF
MJE182     22pF
KSC2331    22pF
KSD1616A   22pF (50C rise in still air)
2N5320     22pF
2SC5706    22pF
2SD1801    22pF
2SD1815    15pF
BD139      15pF
2SD1802    10pF
2SD1803     0pF
2SD1816     0pF
D44C8       0pF (30V peak, reference)
STSA1805    0pF (26V peak)
MJD44H11    0pF (24V peak)
D44H11-ST   0pF (21V peak)
2N3421      0pF (19V peak)

BC639, BD139, and MJE243 I bought off eBay are counterfeit.
(The fake BC639 would work, but I don't use fake parts.)
Notes above are for real parts.

My 2N2219A works, but old typical curves from Motorola
Small-Signal-1984 are unacceptable.  Not recommended.

IN-SYSTEM MEASUREMENTS

Rediscover real FM performance, turn on and turn off.
Maybe the turnoff spike is a bug not a feature.
Maybe we really want to minimize it, not match 091.
Does it affect the in-circuit current ramp if we suppress it with a Miller cap?
Scope output current in 1620, triggered on falling edge of FM control input.
How big/slow before turnon is too slow?  Before the speedup cap is too small?
(A: Speedup cap overwhelms any sane Miller cap.)
Try with KSD1616A no cap, KSD1616A with 450ns worth of cap, 2N3421, and bad 091.

RESULTS:

BC337 with 47pF is non-monotonic
BC337 with 82pF is monotonic, current turnon in 1620 10%-90% is 500ns

In order of ease of installation:

(1) TO-92  STSA1805;
(2) TO-92  BC337, KSD1616A, or MPS651, plus cap;
(3) TO-251 MJD44H11, soldered to copper;
(4) TO-251 2SC5706, 2SC4135, soldered to copper, plus cap;
(5) TO-220 D44C8 bolted to aluminum or copper;
(6) TO-126 MJE182 or BD139 bolted to aluminum or copper, plus cap;
(7) TO-252 2SD1816 etc soldered to PCB
(8) TO-92  Potted in original heat sink.

During testing, I observed that the cap caused VHF oscillation in the 1620.
(It didn't show up on the bench.)  This goes away when you eliminate the
jumper lead inductance.

Fast parts operated fast cause Iout to rise, fall, and rise again, which is bad.
I think it's instructive to the next maintainer to use a fast part with cap
instead of a slow one that happens to be right.  It will make them think.

I think it's not worth it to make a custom circuit board for surface-mount parts.
Fun fact: Nobody can do WJ and FL and FM.  BC337 and MPS651 can do FL and FM.

Except for the Baker Clamp Boys, nobody can do WJ and FL.

2N4401 is one of my recommendations for the 086 in the 1401 AQU Decode Switch card. It can also replace 071 in the MARS/Inhibit cards. Other AQU parts are BC337, KSC1008, and MPS651. I use the MPS651 (with zener) in my 1620 FL cards that lost their 086. It’s a fine part, still cheap and abundant, and it can also be used for the non-saturating Current Driver cards. KSC1008 is fine replacing the Decode Switch 086, but not usable for Current Driver 091 because it exhibits quasi-saturation which renders the Baker clamp ineffective. For my one failed FM card (091), I used a D44C8 for years. They are discontinued now, but a number of still-living parts look fine on datasheet and jig, and a few of those have been confirmed to work perfectly in the 1620, with waveforms almost indistinguishable from 091. (That testing came to a halt when one then another Select line went open-circuit. If they broke at the terminal, it may be possible to repair them. Someday…)

The Current Driver 091 was heavily heat-sinked, but even though Silicon parts run hotter unless you tweak the Baker clamp, they can handle it easily and will probably even survive a fan failure. Even the TO-92’s.

Regards,

Dave Wise