Good transistors for TTL for homemade 4-bit computer

[lordvader88]

One of these days I'd like to make a 4-bit adder/calculator, and I've seen projects that need around 200 transistors. What are good low power transistors for TTL ?

[Ammar]

Someone will certainly jump in and correct me if I am wrong. However, I think TTL is intrinsically not low power. BJTs are known for their higher power consumption, with the trade off being high gain. This is why CMOS took over the digital space. Moderate gain with far lower power consumption beat TTL. If you want to use BJTs, just grap a jelly bean part. If you want low power, use MOSFETs.

[T3sl4co1l]

BJT -- 2N3904, or something smaller like MMBT81, or BFR92 (which is an RF transistor, so use it at low current to save power and avoid oscillation).  Don't expect amazing performance: under 1MHz, probably.

You'll probably do something more like RTL or DTL, because TTL requires more transistors, and you don't really need the drive/fanout characteristics in every single gate.  It's close enough, not a big distinction but probably helpful for researching circuit examples.

NMOS isn't practical with jellybean MOSFETs (i.e., 2N7002), because it's so preposterously slow at any reasonable supply currents (i.e., you won't get into the MHz unless you have 10s of mA to spare).  If you can find a small (like RUM001L02) or RF transistor for as cheap, you'll get reasonable performance.  Otherwise, count on quite slow clock frequencies (like the guy who built the MOS 6502 CPU from 2N7002's: 30kHz max!).

Tim

[BarsMonster]

I spent quite some time researching this topic for the similar project.

If you go for smaller digital circuits (up to 100 transistors) - you could go with FETs and gain experience which is close to real CMOS design.
If you want to go for larger ones - I recommend to go with BJT as they are by far the cheapest. With resistive loading number of transistors will be half that of CMOS. It will also work faster (down to 10ns switching).

For logic part of the circuit I recommend using BC857C/BC847C (or their non-SMD variants) - lowest voltage, highest hFE, almost lowest cost.
2n3904/3906 - are too "big" for logic, you won't need their 200mA collector current, 100mA of BC857C is enough. This 200mA collector current comes with a price of higher capacitance.

For bus drivers I recommend MMBT2222/MMBT2907 (500mA) or 2n3904/3906 (200mA) - but only if you see BC857 is not making it fast enough. Unless you are shooting for maximum speed (>10Mhz) or driving really large buses (like arrays of BJT SRAM) you won't need these.

[lordvader88]

I'm just looking at a basic add/subtract calculator so speed isn't a worry. Like this
http://www.waitingforfriday.com/?p=529

No rush on this project either, and I'm just planning to breadboard it. OK so I'll either look for a good deal on some basic JFETs or just get a a bunch of little BJTs

[rstofer]

Nothing wrong with the BC547 that is used in the project.  I might substitute the venerabe 2N2222A since this is what it was designed for, about 100 years ago (actually 1962).  I would use the metal can version if I could find it.  That is the way it is supposed to be done.  None of this plastic stuff...  Note:  it doesn't really matter but I just like the old metal.

The project itself is RTL - Resistor Transistor Logic.  This was a very common logic family way back when.

https://en.wikipedia.org/wiki/Resistor%E2%80%93transistor_logic

RTL flew to the moon!  It predates DTL (Diode Transistor Logic) and TTL (Transistor-Transistor Logic).

When you design with RTL, you need to think in terms of NOR gates because any input going high will drive the output low.  If any input going high drove the output high, it would be an OR gate.  The inverted output makes it a Not-OR or NOR gate.  Depending on how far you want to go with logic design, you will use DeMorgan's Theorm to show that RTL logic should be written as a product of sums (AND of ORs).  TTL logic is usually designed as a sum of products (OR of ANDs) because TTL is inherently a NAND (Not-AND) configuration.  I'll leave it to you to wander down the rathole of product of sums versus sum of products.  For the project you are copying, such extra credit work is not required.

Looks like a fun project and a great opportunity to play with transistors and logic the way it used to be done!

BTW, RTL was very popular in the hobby community back when I started playing with this stuff in '69.

[rsjsouza]

Nothing wrong with the BC547 that is used in the project.  I might substitute the venerabe 2N2222A since this is what it was designed for, about 100 years ago (actually 1962).  I would use the metal can version if I could find it.  That is the way it is supposed to be done.  None of this plastic stuff...  Note:  it doesn't really matter but I just like the old metal.

You can still find some at BGMicro: http://www.bgmicro.com/TRN2N2222AMetal.aspx

I'm just looking at a basic add/subtract calculator so speed isn't a worry. Like this
http://www.waitingforfriday.com/?p=529

No rush on this project either, and I'm just planning to breadboard it. OK so I'll either look for a good deal on some basic JFETs or just get a a bunch of little BJTs

The 2N2222 is designed for switching and therefore it is a great fit for such project. Obviously it will cost more as it is not mainstream anymore.

[Kleinstein]

If you want speed from jelly bean parts, consider ECL logic as well.
Due to saturation BJT based logic like DTL, RTL is inherently slow unless you have special transistors or add schottky diodes like in the S-TTL series.

[Benta]

+++1000 to ECL.

It has a hell of a lot of advantages:
- Fast
- Constant current consumption
- Low noise (no switching transients)
- Uncritical transistor selection (operate in linear region, so no storage issues)
- Etc.

If you're interested, the "MECL System Design Handbook" (HB205/D) from Motorola, later ON Semi is a must-have.
Unfortunately, ON has  removed it from their website, but I found it on a mirror:

http://d3i5bpxkxvwmz.cloudfront.net/pl/1276642244-MECL-system-design-handbook.pdf

Have fun.

[T3sl4co1l]

I don't recommend ECL.

Not without some experience, first.

It is easy enough to build single gates, but I've found breadboarding discrete ECL leads to oscillators rather than logic functions.  Be careful!

RTL is problem-free, though.  Slow, but easy.

Speaking of saturation: you can make your own half-ass LS TTL by adding a schottky diode from base to collector.  A smallish one like BAT54 or BAT85 is fine.  It will add some B-C capacitance, slowing down the rise/fall, but prevents storage time.

Tim

[rstofer]

The OP wants to build from discretes and ECL will take more than 6 times as many transistors to fabricate a simple gate.  Furthermore, ECL wants to drive a twisted pair transmission line between packages for each signal.  That's a little over the top for an experimental 4 bit adder.  Besides, the reference project is already designed for RTL.

Speed not being a consideration, RTL is just perfect for fooling around.  If it could make it to the Moon, it will probably work on a breadboard.

RTL is as simple as it gets:  One transistor for each gate output and one resistor for each input.  Plus the output resisor.  A 2 input NOR gate takes 3 resistors and 1 transistor.

The reference project doesn't actually use NOR gates, it actually uses AND and OR gates.  These take 3 resistors and two transistors for a 2 input gate.  I'm pretty sure I would back up and look at NOR gates because my output stage doesn't have to pull up.  The pull-up design requires that the base voltage go fairly high and that implies that the input logic level be high as well.  The base has to go higher than the level one output voltage plus the V_BE.  Still, there is no reason it won't work.  To be honest, I would want to build up an AND and OR gate just to see what the switching thresholds look like.

[edavid]

TTL uses gold doped transistors for faster saturated switching speed.

Of course you can't really build discrete TTL, because of the lack of multi-emitter transistors, but if you want to build TTL-like DTL, it makes sense to use gold doped discretes.  They are not as common as they used to be, but PN2369, MMBT2369, and MMBT5179 are still in production.

If you use non-gold-doped transistors like 2N2222 or 2N3904, you'll have to add Baker clamps or Schottky clamps to get any speed.

[Benta]

It is easy enough to build single gates, but I've found breadboarding discrete ECL leads to oscillators rather than logic functions.

Interesting. I've worked a lot with ECL, but only at IC level. The old MC16xx series, then MC10K and MC10H and at the end ECLinPS.
I've never, ever had a problem with oscillations or other bad things.

High frequency CMOS logic was a much bigger problem because of all the switching transients, specially on the supply lines.

Mostly I've done a proper layout with matched lines for prototyping, but breadboarding (with careful floor planning) also works. The plastic blocks where you stick in the parts (don't remember the trade name, never used them) would be a complete disaster.

No matter what, I still recommend HB205/D, the chapter on PCB transmission lines is useful everywhere.

Cheers,

[Benta]

The OP wants to build from discretes and ECL will take more than 6 times as many transistors to fabricate a simple gate.

A 4-input ECL gate needs 6 transistors, a 2-input gate needs 4 (plus bias, but this can be one circuit for the whole board).

Furthermore, ECL wants to drive a twisted pair transmission line between packages for each signal.

ECL works excellently in single-ended configuration, no need for differential signalling.

And just for my curiosity: where do I get twisted-pair PCB traces? My current layout program doesn't support it.

Apart from that: RTL is certainly the simplest solution, and if operating frequency in the 100+ kHz range is enough, go with it.

[rstofer]

The OP wants to build from discretes and ECL will take more than 6 times as many transistors to fabricate a simple gate.

A 4-input ECL gate needs 6 transistors, a 2-input gate needs 4 (plus bias, but this can be one circuit for the whole board).

Furthermore, ECL wants to drive a twisted pair transmission line between packages for each signal.

ECL works excellently in single-ended configuration, no need for differential signalling.

And just for my curiosity: where do I get twisted-pair PCB traces? My current layout program doesn't support it.

Apart from that: RTL is certainly the simplest solution, and if operating frequency in the 100+ kHz range is enough, go with it.

The only time I have seen ECL used, it was wire-wrapped with twisted pairs.  That's why there is a true and complementary output.  Most packaged logic has only a single ended input and differential signaling isn't required.  This does not preclude the possibility of a differential input on custom logic where twisted pair is used everywhere.  Nevertheless, ECL simply isn't applicable to a first time project using discretes.

I built up a 2-input AND followed by a 1 input OR (I wanted the emitter resistor so I didn't just use an inverter) and, sure enough, there will be a limit to the number of inputs in the AND configuration.  The voltage drops across the series transistors gets out of hand fairly quick.  The OR gate straightens everything out.  The waveform looks good at 100 kHz and even up to 500 kHz but at 1 MHz the waveform isn't very good.  Since we aren't talking about clocked logic, not much of this matters.  The project is designed to run at toggle switch rates.

I have attached a scope image of the 100 kHz input (cyan) and output (yellow).  There is a large delay in rise and fall times.