27 Oct
2025
27 Oct
'25
3:26 a.m.
It seems that nowadays you can't get *any* replacement for failing DEC
3639 aka 2N3639 transistors. All parts are obsolete and unobtanium, e.g.
2N3640, PN3640, MMBT3640 and so on.
So, what can be used instead? The most important electrical parameter is
the storage time. It needs to be *very* low, around 20-30 ns.
Does it mean that a failing PDP-8 will stay a dead PDP-8 from now on?
Christian
27 Oct
27 Oct
5:41 a.m.
https://www.ebay.com/itm/334846296489
2N3639 - Silicon PNP Transistor MFG -FAIRCHILD
$5, more than 10 available
or do you know them to be a problem?
7 a.m.
On Oct 27, 2025, at 6:26 AM, Christian Corti via
cctalk <cctalk(a)classiccmp.org> wrote:
It seems that nowadays you can't get *any* replacement for failing DEC 3639 aka
2N3639 transistors. All parts are obsolete and unobtanium, e.g. 2N3640, PN3640, MMBT3640
and so on.
So, what can be used instead? The most important electrical parameter is the storage
time. It needs to be *very* low, around 20-30 ns.
Does it mean that a failing PDP-8 will stay a dead PDP-8 from now on?
Christian
Have you tried with an ordinary switching transistor that's readily available?
I'm surprised at that storage time, why would it need to be that low for a machine
that runs at just a couple of MHz?
paull
9:37 a.m.
On Mon, 27 Oct 2025, Paul Koning wrote:
Have you tried with an ordinary switching transistor
that's readily
available? I'm surprised at that storage time, why would it need to be
that low for a machine that runs at just a couple of MHz?
The storage time is very crucial. We have tested many different switching
transistors. The point is, the PDP-8 uses needle impulses (around 30ns) to
switch a flip-flop. They just don't work with today's transistors. And
on modules like the R211/S211, they introduce a too long delay.
Christian
12:17 p.m.
On 2025Oct 27,, at 9:37 AM, Christian Corti via cctalk <cctalk(a)classiccmp.org>
wrote:
On Mon, 27 Oct 2025, Paul Koning wrote:
Cob, fT not quite as good, but close:
2N5400: Cob=6p0, fT=400MHz
2N4916: Cob=4p5, fT=400MHz
vs 2N3639: Cob=3p5, fT=500MHz
Perhaps amongst those you tried (?).
I see from the schematic they were going to quite the effort to maintain speed while still
using capacitive triggering.
Have you tried with an ordinary switching
transistor that's readily available? I'm surprised at that storage time, why
would it need to be that low for a machine that runs at just a couple of MHz?
The storage time is very crucial. We have tested many different switching transistors.
The point is, the PDP-8 uses needle impulses (around 30ns) to switch a flip-flop. They
just don't work with today's transistors. And on modules like the R211/S211, they
introduce a too long delay. 
2:53 p.m.
On 10/27/25 14:17, Brent Hilpert via cctalk wrote:
On 2025Oct 27,, at 9:37 AM, Christian Corti via cctalk
<cctalk(a)classiccmp.org> wrote:
The MPSH81 shows Ccb = 0.85 pF and Cce = 0.65 pF and fT 600
MHz. They don't give a Cob spec.
Jon
On Mon, 27 Oct 2025, Paul Koning wrote:
Cob, fT not quite as good, but close:
2N5400: Cob=6p0, fT=400MHz
2N4916: Cob=4p5, fT=400MHz
vs 2N3639: Cob=3p5, fT=500MHz
Have you tried with an ordinary switching
transistor that's readily available? I'm surprised at that storage time, why
would it need to be that low for a machine that runs at just a couple of MHz?
The
storage time is very crucial. We have tested many different switching transistors. The
point is, the PDP-8 uses needle impulses (around 30ns) to switch a flip-flop. They just
don't work with today's transistors. And on modules like the R211/S211, they
introduce a too long delay. 
28 Oct
28 Oct
12:35 a.m.
Den mån 27 okt. 2025 kl 23:03 skrev Jon Elson via cctalk <
cctalk(a)classiccmp.org>gt;:
On 10/27/25 14:17, Brent Hilpert via cctalk wrote:
They actually show the output capacitance in a diagram as a function of
reverse bias voltage. Between 0.8 pF and 1.8pF depending on voltage.
/Mattis
On 2025Oct 27,, at 9:37 AM, Christian Corti via
cctalk <
cctalk(a)classiccmp.org> wrote:
> On Mon, 27 Oct 2025, Paul Koning wrote:
>> Have you tried with an ordinary switching transistor that's readily
available? I'm surprised at that storage time, why would it need to be
that low for a machine that runs at just a couple of MHz?
> The storage time is very crucial. We have
tested many different
switching transistors. The point is, the PDP-8 uses needle impulses (around
30ns) to switch a flip-flop. They just don't work with today's transistors.
And on modules like the R211/S211, they introduce a too long delay.
Cob, fT not quite as good, but close:
2N5400: Cob=6p0, fT=400MHz
2N4916: Cob=4p5, fT=400MHz
vs 2N3639: Cob=3p5, fT=500MHz
The MPSH81 shows Ccb = 0.85 pF and Cce = 0.65 pF and fT 600
MHz. They don't give a Cob spec.
Jon
27 Oct
27 Oct
1:14 p.m.
On 2025-10-27 8:00 a.m., Paul Koning via cctalk wrote:
The 20 Mhz time state generator in some PDP 8's just kills that idea.
Ben.
On Oct 27, 2025, at 6:26 AM, Christian Corti via
cctalk <cctalk(a)classiccmp.org> wrote:
It seems that nowadays you can't get *any* replacement for failing DEC 3639 aka
2N3639 transistors. All parts are obsolete and unobtanium, e.g. 2N3640, PN3640, MMBT3640
and so on.
So, what can be used instead? The most important electrical parameter is the storage
time. It needs to be *very* low, around 20-30 ns.
Does it mean that a failing PDP-8 will stay a dead PDP-8 from now on?
Christian
Have you tried with an ordinary switching transistor that's readily available?
I'm surprised at that storage time, why would it need to be that low for a machine
that runs at just a couple of MHz?
paull
8:16 a.m.
On 10/27/25 05:26, Christian Corti via cctalk wrote:
It seems that nowadays you can't get *any*
replacement for
failing DEC 3639 aka 2N3639 transistors. All parts are
obsolete and unobtanium, e.g. 2N3640, PN3640, MMBT3640 and
so on.
So, what can be used instead? The most important
electrical parameter is the storage time. It needs to be
*very* low, around 20-30 ns.
I saw a data sheet that mentioned an MPS3639. But, I didn't
find a distributor that had these, either.
Possibly not a direct substitute due to pinout, but the
MPSH81 is a very fast, low-capacitance PNP transistor. I
have used it to replace germanium PNP transistors in some
gear. Just a total drop-in replacement, didn't need to
touch the bias network.
Jon
8:32 a.m.
On 10/27/25 05:26, Christian Corti via cctalk wrote:
It seems that nowadays you can't get *any*
replacement for
failing DEC 3639 aka 2N3639 transistors. All parts are
obsolete and unobtanium, e.g. 2N3640, PN3640, MMBT3640 and
so on.
So, what can be used instead? The most important
electrical parameter is the storage time. It needs to be
*very* low, around 20-30 ns.
Oh, it apparently is Silicon. So, an MPSH81 should really
do well.
Jon
11:46 a.m.
On 10/27/25 11:41, Christian Corti via cctalk wrote:
On Mon, 27 Oct 2025, Jon Elson wrote:
I have used its NPN mirror (MPSH10) in switching circuits
and they work quite well, giving ~1ns 10-90% rise and fall.
I have not used the MPSH81 quite as much. I most recently
use the MMBTH81 and MMBTH10 in the SO23 package.
I think you really ought to try one and see how it works.
Obsolete is not a big hurdle, several distributors usually
have a bunch of these parts.
Jon
Oh, it apparently is Silicon. So, an MPSH81
should
really do well.
I think it won't. It is obsolete and isn't sold any more.
And especially, it is not a switching transistor, so it
will perform very poorly in a digital circuit.
28 Oct
28 Oct
1:30 a.m.
On Mon, 27 Oct 2025, Jon Elson wrote:
I have used its NPN mirror (MPSH10) in switching
circuits and they work
quite well, giving ~1ns 10-90% rise and fall. I have not used the MPSH81
Rise/fall times are not the problem. The problem is the huge delay of over
100ns due to the storage time.
Christian
12:03 p.m.
Schotky diode the keep the base out of saturation ?
As used in 74S 74LS, appreciate it's an extra component to bodge in
see eg
https://electronics.stackexchange.com/questions/242715/using-a-schottky-dio…
Martin
-----Original Message-----
From: Christian Corti via cctalk <cctalk(a)classiccmp.org>
Sent: 28 October 2025 08:30
To: Jon Elson via cctalk <cctalk(a)classiccmp.org>
Cc: Christian Corti <cc(a)informatik.uni-stuttgart.de>
Subject: [cctalk] Re: Replacements for DEC 3639
On Mon, 27 Oct 2025, Jon Elson wrote:
I have used its NPN mirror (MPSH10) in switching
circuits and they
work quite well, giving ~1ns 10-90% rise and fall. I have not used
the MPSH81
Rise/fall times are not the problem. The problem is the huge delay of over 100ns due to
the storage time.
Christian
12:58 p.m.
On 10/28/25 14:03, Martin Bishop via cctalk wrote:
Schotky diode the keep the base out of saturation ?
As used in 74S 74LS, appreciate it's an extra component to bodge in
see eg
https://electronics.stackexchange.com/questions/242715/using-a-schottky-dio…
That's called a Baker clamp, was commonly used back in the
mid '60s.
Jon
1:21 p.m.
I did wonder if the DEC 3639 part might have an integral Schotky diode, a question for DEC
..., a bit unobtanium these days but bipolar transistor + Schotky diode do / did exist -
as integrated components.
Martin
-----Original Message-----
From: Jon Elson via cctalk <cctalk(a)classiccmp.org>
Sent: 28 October 2025 19:59
To: cctalk(a)classiccmp.org
Cc: Jon Elson <elson(a)pico-systems.com>
Subject: [cctalk] Re: Replacements for DEC 3639
On 10/28/25 14:03, Martin Bishop via cctalk wrote:
Schotky diode the keep the base out of saturation ?
As used in 74S 74LS, appreciate it's an extra component to bodge in
see eg
https://electronics.stackexchange.com/questions/242715/using-a-schottk
y-diode-to-improve-pnp-transistor-turn-off-time
That's called a Baker clamp, was commonly used back in the mid '60s.
Jon
1:28 p.m.
On Tue, 2025-10-28 at 14:58 -0500, Jon Elson via cctalk wrote:
On 10/28/25 14:03, Martin Bishop via cctalk wrote:
IBM 1401 SMS circuits used an inductor, typically 56µh but sometimes as
small as 1.5µh, which I was told was to reduce either the switching or
storage time (I don't remember which). See for example
http://vandyke.mynetgear.com/1401/docs/ALD/SMS/370143/370143.pdf or http://vandyke.mynetgear.com/1401/docs/ALD/IBM-StandardModularSystem-Neff7.…
or http://vandyke.mynetgear.com/1401/docs/ALD/223-6875-1_IBM-SMS.pdf
Is there a name for that?
Schotky diode the keep the base out of saturation
?
As used in 74S 74LS, appreciate it's an extra component to bodge in
see eg
https://electronics.stackexchange.com/questions/242715/using-a-schottky-dio…
That's called a Baker clamp, was commonly used back in the
mid '60s.
29 Oct
29 Oct
2:56 a.m.
On Tue, 28 Oct 2025, Jon Elson wrote:
On 10/28/25 14:03, Martin Bishop via cctalk wrote:
That is a nice idea. We will try that and find out if this helps. I am
also thinking of using *cough* P-channel MOSFETs like BS250 instead.
There must be a solution with available parts ;-)
Christian
Schotky diode the keep the base out of saturation
?
As used in 74S 74LS, appreciate it's an extra component to bodge in
see eg
https://electronics.stackexchange.com/questions/242715/using-a-schottky-dio…
That's called a Baker clamp, was commonly used back in the mid '60s. 
11:38 a.m.
Hi Christian,
what do You think about:
2N3640 -> 2N3906
2N3693 -> 2N3904, 2N2222A, BC337-25
Best
Andreas
Please contact me offlist.
Am 27.10.2025 um 16:38 schrieb Jon Elson via cctalk
<cctalk(a)classiccmp.org>rg>:
On 10/27/25 05:26, Christian Corti via cctalk wrote:
It seems that nowadays you can't get *any*
replacement for failing DEC 3639 aka 2N3639 transistors. All parts are obsolete and
unobtanium, e.g. 2N3640, PN3640, MMBT3640 and so on.
So, what can be used instead? The most important electrical parameter is the storage
time. It needs to be *very* low, around 20-30 ns.
Oh, it apparently is Silicon. So, an MPSH81 should really do well.
Jon
30 Oct
30 Oct
1:30 a.m.
On Wed, 29 Oct 2025, hupfadekroua wrote:
what do You think about:
2N3640 -> 2N3906
2N3693 -> 2N3904, 2N2222A, BC337-25
I've tested exactly those types. They are all too slow. For example, when
you cascade the three independent stages on an R211 module, feed the first
stage with a clock of around 1 MHz, and just replace one of the 2N3639
with a 2N3906, the turn-off slope after the last stage will be delayed
from around 30ns to 200ns!
But we will test the trick with the Baker clamp on the 2N3906.
Christian
4 a.m.
The European equivalent for the 2N3639 / 2N3640 is BSX36
And for the 2N3693 is BF365, BF241 and also 2N3664
This according the 'DÖV vergleichstabelle' maybe that's of any help.
I didn't check the availability of the parts, but a quick ebay check
If it's of any help I do have about 10pcs 2N3639 (NOS) in stock.
-Rik
-----Oorspronkelijk bericht-----
Van: Christian Corti via cctalk <cctalk(a)classiccmp.org>
Verzonden: donderdag 30 oktober 2025 09:31
Aan: hupfadekroua via cctalk <cctalk(a)classiccmp.org>
CC: Christian Corti <cc(a)informatik.uni-stuttgart.de>
Onderwerp: [cctalk] Re: Replacements for DEC 3639
On Wed, 29 Oct 2025, hupfadekroua wrote:
what do You think about:
2N3640 -> 2N3906
2N3693 -> 2N3904, 2N2222A, BC337-25
I've tested exactly those types. They are all too slow. For example, when
you cascade the three independent stages on an R211 module, feed the first
stage with a clock of around 1 MHz, and just replace one of the 2N3639 with
a 2N3906, the turn-off slope after the last stage will be delayed from
around 30ns to 200ns!
But we will test the trick with the Baker clamp on the 2N3906.
Christian
31 Oct
31 Oct
1:21 a.m.
On Thu, 30 Oct 2025, hp-fix(a)xs4all.nl wrote:
The European equivalent for the 2N3639 / 2N3640 is
BSX36
I've been through all "Vergleichstabellen" etc. Alas, the BSX36 is also
unobtanium.
If it's of any help I do have about 10pcs 2N3639
(NOS) in stock.
I'll try the Baker clamp first, but I need to order a batch of BAT46,
first. I'll also order a couple of BS250 just to test the MOSFET idea.
Christian
30 Oct
30 Oct
8:41 a.m.
On 10/30/25 03:30, Christian Corti via cctalk wrote:
On Wed, 29 Oct 2025, hupfadekroua wrote:
I think you should try the MPSH81. Depending on the bias
and base drive, you may also need the Baker clamp. (Use a
really fast, low capacitance diode for that.)
The 2N3906 is an audio transistor, and not fast or low
capacitance.
Jon
what do You think about:
2N3640 -> 2N3906
2N3693 -> 2N3904, 2N2222A, BC337-25
I've tested exactly those types. They are all too slow.
For example, when you cascade the three independent stages
on an R211 module, feed the first stage with a clock of
around 1 MHz, and just replace one of the 2N3639 with a
2N3906, the turn-off slope after the last stage will be
delayed from around 30ns to 200ns!
But we will test the trick with the Baker clamp on the
2N3906. 
27 Oct
27 Oct
9:22 p.m.
On ETSY there were 24 pieces of the MPS3639 for a very reasonable price.
They sold a few hours ago, so Christian may have found what he was looking
for.
On Mon, Oct 27, 2025 at 6:33 PM Christian Corti via cctalk <
cctalk(a)classiccmp.org> wrote:
It seems that nowadays you can't get *any*
replacement for failing DEC
3639 aka 2N3639 transistors. All parts are obsolete and unobtanium, e.g.
2N3640, PN3640, MMBT3640 and so on.
So, what can be used instead? The most important electrical parameter is
the storage time. It needs to be *very* low, around 20-30 ns.
Does it mean that a failing PDP-8 will stay a dead PDP-8 from now on?
Christian
28 Oct
28 Oct
1:33 a.m.
On Tue, 28 Oct 2025, Tom Hunter wrote:
On ETSY there were 24 pieces of the MPS3639 for a very
reasonable price.
They sold a few hours ago, so Christian may have found what he was looking
for.
It wasn't me. And I need more than just 24. And I can't order in the US.
Christian
0
days inactive
4
days old
24 comments
12 participants
participants (12)
-
ben -
Brent Hilpert -
Christian Corti -
hp-fix＠xs4all.nl -
hupfadekroua -
Jon Elson -
Martin Bishop -
Mattis Lind -
Paul Koning -
StricK -
Tom Hunter -
Van Snyder