Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Germanium transistors: logic circuits in the IBM 1401 computer (righto.com)
85 points by picture on March 3, 2021 | hide | past | favorite | 55 comments


From what i have read, with my limited understanding of chipmaking, it seems that UO2(Uranium dioxide) makes a great heat and radiation resistant semiconductor with better properties than silicon or germanium[1]. So how come we arent using that or exploring other materials for pushing the limits of computational power?

[1]: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.52...


Probably a few factors:

- Difficulty growing high quality crystals

- All the process chemistry for building devices would be different

- Toxicity/radioactivity

The third one is IMO the least important. There are a lot of semiconductor materials that theoretically have properties comparable to or better than silicon for solar cells (as this paper suggests). But having a theoretically good band gap doesn't matter if the actual material as grown is full of deleterious defects. We're really good at growing near-perfect silicon crystals. We're not so good at growing perfect crystals of iron (II) sulfide, uranium dioxide, etc.

Even if that could be solved, re-developing all the chemistry and tooling for fabricating dense logic/memory devices based on uranium dioxide instead of silicon would require an extraordinary investment of money and time. It could be comparable to or even more expensive than the transition to extreme ultraviolet lithography.


Chemical scarcity should be in your list, and it is the reason silicon will remain for years to come, at least in electric domains.


Yep. I tried to ballpark the amount of polysilicon used to actually make ICs. Somewhere between 100,000 and 10 million tonnes a year, I figure. Those little dies are only 100 mg or so but we print them by the billions. Whether the real number is more to the high or low end of that, it's hopeless. As Wikipedia says "About 118 tonnes of germanium was produced in 2011 worldwide."


It can require very pure materials to make conductors and semiconductors or they won't act right.

And it can take kilos of 99 percent pure raw material to end up with a few grams of 99.999 or better.

Silicon was already available in rail cars having 99+ percent purity for steelmaking.


U238 is an alpha emitter. You'd probably get way more bitflips than non-radioactive semiconductors experience.


The mere mention of uranium is going to make the paranoid scared, and nuclear regulatory agencies raise much concern.

Even "depleted" uranium is somewhat radioactive, and it's a toxic heavy metal.


A modern pc is toxic, it's not like you could survive stuffing an iPad in a blender and eating it. Thats why e-waste is causing so many problems.


I doubt an iPad would kill you. One guy ate 7 TV sets, a computer, bicycles, and an entire Cessna 150 aircraft.

https://en.wikipedia.org/wiki/Michel_Lotito


The Frenchman ate an airplane and died of natural causes at the age of 57.

Now there's a sentence that I never thought I'd write.


> it's not like you could survive stuffing an iPad in a blender and eating it

To be fair, I think the sharp chunks of metal and plastic would kill you before the heavy-metal poisoning could.


Lots of stuff far worse than uranium is used in semiconductor manufacturing.


Not specifically about the 1401, but where do you find the time to produce this volume of high-quality content that obviously takes a ton of work and research?


The secret is to retire :-)


Ken Shirriff, your blog is just awesome.

What a wealth of material you have on righto.com!

Blown away.


Cripes. I just checked my reading backlog file (well, the programming-oriented file) and it shows 12 righto posts


Neat!

NOS germanium transistors are still highly sought after(and therefore expensive) for music gear such as guitar distortion pedals and pre-amps. They impart a lovely ‘gooey’ quality that modern transistors just can’t do.


Yep, I have a diy fuzz effect pedal around here somewhere that uses a matched pair of germanium transistors. An older electrical engineer gave me a box of them after I explained how sought after they were. He couldn't get his head around the idea that I would want to use them in a new project when new silicon transistors would be superior for many reasons.


It seems like audio applications like to use old parts: how do germanium transistors compare with tubes?


I would say they are like tubes because the distortion is more interesting than the silicon alternatives.

For non-distorted applications if you do overdrive a bit it can be a lot less harsh too.

Also very difficult to make two the same, so specs were wide and top parts were hand selected from large batches of on-spec parts.

One of my best sources of tasty germaniums was from boards about the same vintage as these IBMs.

I think it's good to make musical or audio circuits from just about any tube or transistor whether it was originally intended for audio or not.

A very simple 2-transistor PCB can have art where it can mirrored and used for NPN or PNP, with positive or negative ground to your battery clip.

I like transistor sockets like there were on some of the early solid-state scientific instruments. Before they could be sure the transistors were more reliable than the sockets.

Put on adjustable bias and find some superb high-performance small-signal BJT silicons for reference. Leave one in then audition the germaniums in the other socket, biasing each one accordingly.

Without a guitar pedal enclosure you can just cut a long guitar cable in half and solder it to the PCB's input & output with the battery hanging off. If the situation arises the whole thing could then be heat-shrunk right there into the middle of the cable. Don't ask me how I know. The python that swallowed an alligator. Doesn't matter if it's positive or negative ground since it's floating.

One megohm impedance for guitar input is good, and play it by ear for a wide choice of output impedance into the guitar amplifier.

I guess I kind of like playing a fuzzy guitar as much as ZZTop :)

Been known to make a pedal or two.


Depends on the use. For low distortion, tubes fare better because their operating point is more linear than any single transistor. Ge doesn't make a huge difference here iirc.

Outside their operating point is another story. As said elsewhere, Ge has a smaller and less sharp drop than Si, which is worlds different for all sorts of clipping distortion applications. The classic example is the Fuzz Face, where the gentler clip made for a warmer sound. Think early Hendrix versus late Hendrix. But the transistors were so inconsistent and the circuit so sensitive to the gain that only one in fifty would sound good!


Tubes are better since they have amplify with less "zero point crossover distortion". Silicon has .7v and Germanium has .3v Vbe so that is why people seek them.

My understanding though is that it is something you can tune by adding biasing diodes but requires matching.

I am not an expert...just something I remembered from school :)

https://en.wikipedia.org/wiki/Crossover_distortion

Edit: added Vbe


Crossover distortion is completely inaudible if the amplifier is correctly biased as it amounts to a fraction of a fraction percent of the output power, like attempting to listen to a distorted 100 mW pocket radio put on top of a perfectly linear 500W amplifier set at maximum output. The difference between germanium and silicon transistors lays in the former having much narrower bandwidth, so that it naturally and more gracefully limits higher audio frequencies, and some the artifacts of distortion become less audible making the sound more warm. Tubes are better than both in this context because they saturate in a really graceful way and produce mostly even harmonics which give tube guitar amps their unique different sound; this is something hardly reproducible in solid state, unless DSPs are involved. I wouldn't however use tubes for clean HiFi amplifiers as more modern technologies such as Class D today made possible building incredibly good amplifiers at a fraction of the cost, space and energy requirements. They can also be used for guitars, but as Class D stages distortion sounds like pure junk, everything has to be redesigned so that the final stages are kept clean and distortion happens before entering them.


The distortion introduced by particular electronics is considered an inherent part of the instrument's musical quality by some musicians. Swapping out a high fidelity modern class D amplifier for an old tube amp is kind of like putting steel strings on a classic guitar or violin, instead of catgut or at least a modern plastic meant to emulate catgut. The instrument would never have sounded that brilliant to the musician who wrote a piece before steel strings were introduced.

From an engineering perspective, guitar tube amplifiers are horrible at fidelity. But in terms of accurately reproducing what 1950s rock-and-roll sounded like live, they're ideal.


Semi-related, Brian Wampler made this[1] video where he tries a bunch of different opamps and diodes in some overdrive/distortion circuits, including some germanium and various colored LEDs for good measure.

Not super-scientific but good fun.

[1]: https://www.youtube.com/watch?v=PH2uj9bWfM4


Glad to see other people that enjoy building guitar pedals. That seems increasingly rare. There are some interesting read if you like electronics and guitars:

https://www.analogman.com/fuzzface.htm

http://www.geofex.com/article_folders/fuzzface/fftech.htm

Another one

https://jacquespedals.com/the-ac128-transistor/

An excerpt:

"EPILOGUE : When supply will dry…

In a very close future, all stocks, including mine, of original germanium AC128 or any usable germanium transistor will dry.

Forever.

After this extinction, that only a few aficionados will notice, there will be no other possiblities to obtain THE original fuzz tone than a used Germanium fuzz box or a digital model."


A problem with germanium transistors is their tendency to degrade over time until they either start leaking too much current also when unbiased or go directly to short circuit, no matter if they're used or not, so it's advisable to test a small stock before buying bigger lots that could contain many now defective parts. Speaking from experience as I had to ditch like 25% of a big lot purchased years ago in which many were shorted to death.

About the AC128, I stopped using it in pedals years ago. It was an ordinary low cost part that later became famous for being used in the original fuzz pedal; as a result it is today sold at outrageous prices, while any normal Ge transistor would work with no difference in sound. It's not the part number, it's the technology: Ge transistors are slower (as in narrower bandwidth) and their sound is therefore less harsh, warmer than Si ones. A pair of ultra cheap ACY* SFT* or many russian ones (less prone to leaks due both to higher [military] standards and their relatively shorter age as the soviets kept producing them long after the west stopped) would sound equally great once the circuit is being adjusted for bias and gain.


Author here for your IBM 1401 questions :-)


More an IBM archeology question: aside from the scattered projects to catalog specific product lines, are you aware of a more general effort? I've been pulling together a lot of their research papers and documentation related to the POWER architecture for a while, and I've noticed some pretty big missing pieces. For example: in order to best preserve the pdfs, while increasing their usefulness and reducing their maintenance burden (disk space), I set out to convert them to PDF/A. That means embedded fonts, which is fine - I'd ideally digitize as much of the text as possible (instead of throwing a hidden OCR layer on and calling it a day), but... the font is nowhere to be found. IBM had an extremely popular font, Press Roman, that is very common in journals and books published from the 60s to the late 80s, and as far as I can tell - it may now exist only in whatever remains of two printer font cartridge SKUs. I won't even get started on the problem regarding their plotting software.

It reminds me of how NASA simply lost so much of the original media, and what he have today is either purely accidental or the result of a considerable amount of work done by volunteer restoration groups. We really need to get a grip on this problem now, and it would be nice if IBM would actually help out with that - instead of leaving it to volunteers.


How similar or different are the performance characteristics of the NPN-based and PNP-based gates? Assuming there's a meaningful difference, are there signs in the design anywhere where this was considered? Thinking of cases where complex logic is pushed to the NPN stages and the PNP stage is used for just an inverter, or similar patterns.


That's an interesting question. Looking at the documentation the NPN card (CHWW) and the PNP card (CGWW) have different switching characteristics (NPN is faster on and slower off). But I haven't noticed any difference in usage patterns.


My memory...which is very hazy and limited on this topic...is that the Germanium based transistors have a VBE of .3v instead of silicon's VBE of .7v

So in things like amplifiers it gives you less zero point crossover distortion. In digital I believe it lends to a faster switch time but I could be very wrong on this part.

Germanium is much more rare so the cost is naturally higher...from what I know.


What's DEFL - Diode Emitter Follower Logic, suppose it's IBM's name for ECL? Do we have examples of these circuits being used in critical circuits in mainframes of the 60s?

I can't imagine a logic circuit of that era that was technically capable of being clocked at 100Mhz, 30 years before these frequencies became the norm.


I've looked, but I can't find anything more about DEFL. (The table says it was used in certain serial numbers of the core memory for the Stretch computer, so it seems rather obscure.)

Based on the speed, it's plausible that it is a type of ECL, but I don't know where the diodes would fit in. IBM had other names such as current-steering logic for ECL.

Keep in mind that the 10ns speed doesn't mean you can run at 100 Mhz, since there are likely to be multiple levels of logic, as well as other delays. But it's still pretty fast.


Is it possible this is DTL (Diode Transistor Logic) in IBM-speak?


It's possible. IBM had several other types of diode-transistor logic including DDTL, SDTRL and CTDL. The emitter follower was more often used with ECL.


I found an IBM patent for high-speed logic using diodes and an emitter-follower. I don't know if this is "Diode Emitter Follower Logic" but it kind of matches the name.

https://patents.google.com/patent/US3050641A/


Off topic: I understand that William Shockley was interested in exploring germanium semiconductors, which may have been a contributing factor in the exodus of eight of his senior scientists and engineers to form Fairchild Semiconductor. Could you comment?


I haven't heard anything about Shockley wanting to explore germanium. My understanding is the "Traitorous Eight" left both because of personality conflicts with Shockley and because he was focusing too much on his 4-layer diode instead of transistors.


Seymour Cray expressed interest in germanium transistors in some speeches I have seen on Youtube. Mainly because of switching speed of germanium


Can I ask a really basic question?

I understand that transistors amplify current but how do they amplify voltage (like in CMOS)? Isn't voltage going to amplify current also?

I never understood this.


I'm not sure I can answer that without confusing things more, but I'll see what I can do. In an MOS transistor, the gate is insulated, so there is no base current, just voltage. This voltage controls the transistor.

In a bipolar (NPN or PNP) transistor, the current through the base causes a larger current through the collector, amplifying by the beta factor. So the transistor is amplifying current. But the current depends on the voltage between the base and emitter, so from that perspective the voltage controls the transistor too.

Whether you're amplifying current or voltage depends on the circuit, so I can't give more than a handwaving answer.


I will also try.

In field effect transistor (in which the actual physics involved are at least for me simpler to grasp) the gate is isolated from everything else and voltage at the base directly changes the geometry of the conductive channel between the S and D pins. In effect the gate voltage directly influences the resistance of the component. MOS is an name for particular practical realization of this mental model.

In bipolar junction transistor (ie. PNP/NPN) there are two diodes that are positioned just so that conduction of one of them influences the other in such a way that when one is positively biased the other will conduct even when reverse biased. For the typical BJT these two diodes have significantly different construction and thus there is difference between emittor and collector, but the effect works both ways (and in fact many circuits will somewhat work even with the 2N3904 connected the wrong way around). The effect is also caused by change of properties of doped semiconductor material in response to electric field gradient but (at least for me) there is no directly applicable model involving discrete lumped components changing their parameters in response to external stimuli that matches the underlying physical principle.


> I understand that transistors amplify current but how do they amplify voltage

Any voltage through a resistor will produce a current, and any current through a resistor will produce a voltage. By properly connecting resistors at the transistor base and collector, one can turn a driving voltage into a current and the collector current back into an output voltage. The basic common emitter transistor amplifier circuit is a good example as it shows how a current amplifier like a transistor is used to amplify a voltage. Resistors are the secret that allow all permutations: voltage to voltage, current to voltage, voltage to current, current to current.


Transistors change their effective resistance based on the "control stimulus", which in turn changes how much current can pass through them.

In BJT transistors the "control stimulus" is the current flowing through the base pin, while for (MOS)FET transistors it's a voltage potential between the gate and source pins.

The amplification happens because in the right region the change in effective resistance is high for small variations in the "control stimulus".

If you drive a BJT with a resistor in front of the base pin, you can drive it with a voltage. If you put a resistor between the gate and source pins, you can drive a MOSFET with a current source.

So the way you control them is different, but what they end up doing is the same. And by using a resistor you can effectively change the way you drive them.


Since Vbe = const the transistor alone cannot amplify voltage. You need a resistor to convert the input voltage into current.


This is exactly it.

Your DC battery or power supply provides your headroom, and the transistor Base or Gate senses a small increment of that voltage and can sometimes deliver as much current as it takes to push the voltage across a resistance or impedance right up to the rails.


The photo of the card drawer open is really spectacular. I am curious how they identified faults with that number of individual cards. Any insights into that?


Having tracked down a few problems in the IBM 1401, it's basically a matter of figuring out what the overall problem is, and then looking for a signal that indicates the problem. Then you work backwards through the circuitry, probing signals with an oscilloscope until you find something that might be the root of the problem. Then you can swap the potentially-bad card and hopefully it fixes the issue.

Of course, some times this is easy, and some times it is hard. Once you find the bad card, the restoration team likes to fix the bad card rather than just replace it. This can have its own challenges. I spent a long time trying to fix a card from the printer that ended up having a cracked trace that was intermittently bad. All the signals looked good, yet the card didn't work.

For a detailed look at a problem, I wrote a blog post about a core memory problem that turned out to be an inductor that failed open. After fixing that, the computer wouldn't power on. We replaced some weak transistors in the power supply, but that didn't help. Finally we found an undocumented fuse that had blown, probably from one of the boards we swapped. Replacing that got everything running again.

http://www.righto.com/2017/12/repairing-1960s-mainframe-fixi...


"restoration team" sounds like everything I like about old computers and general science fiction. Like a reference to Doctor Who or Philip K. Dick.

Anyway, thank you for those detailed information on the article.


Wow this link is fascinating! How fun that you got to be a part of that. I'm trying to imagine an IBM field tech back in the day being called to a customer site and having to go through similar with a customer who was renting one of these for exorbitant sums. It must have been stressful. I'm sure there was some sort of SLA.


I've talked to old-time IBM service engineers and IBM would do just about anything to keep a customer's computer running. An engineer could call his boss and say that a customer needed a very expensive replacement core memory, and IBM would send someone on the next flight to deliver it. If the customer was in the middle of nowhere, someone would drive out with a part and the engineer would drive to meet them halfway. There's a reason people say that nobody ever got fired for buying IBM.


IBM is the only company that will do a proper service contract with folks in rural areas. Heck, the IBM service tech arrived at our door with a new RAID card for our machine running IBM i before I knew there was a problem.


Thanks for yet another great article! I was wondering about the inductor connected to the collector but you already answered in the footnote.


Funny, I was just over on ibm1130.org the other day - so the ALD diagram jumped right out.

http://media.ibm1130.org/1132_Printer_ALD_Jun67.pdf




Consider applying for YC's Summer 2026 batch! Applications are open till May 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: