Megaprocessor - bloke makes 16 bit computer from 14,000 discrete transistors

[Howardlong]

Apologies if this has already been covered, it made El Reg a day or so ago.

In summary, an FPGA engineer is making a complete 16 bit computer from 14,000 discrete transistors. Quite a project.

http://megaprocessor.com/index.html

http://www.theregister.co.uk/2015/06/23/brit_mega_processor/

[Mechanical Menace]

14,000 discrete transistors and counting...

I've started similar but admittedly much, much less ambitious projects myself and always get to the point where I think "Well we know how to build gates using transistors what's the point in not using logic IC's? And now we can make a carry lookahead adder, get that to subtract etc why not use an MPU..."

Thanks for sharing this, I'll definitely follow any progress.

[John Coloccia]

I'll be impressed when he can do it with nothing but batteries, like that other guy that was posting here a couple of weeks ago.  (edit: dear oh dear...I tried to find what happened to the battery idea, and he appears to have had a meltdown  )

Seriously, it's a very cool project, made all the cooler by how utterly impractical and more or less worthless it is!  The ultimate doing just for the sake of doing.    I hope he finishes it and it gets displayed and used somewhere to do some other utterly impractical and worthless thing...like, maybe program it to run a gigantic and complicated Useless Box, and stick it in science museum for everyone to marvel at.

[GK]

I have to get my EDUC16 website started.

[JoeN]

Why ? - long answer : Computers are quite opaque, looking at them it's impossible to see how they work. What I would like to do is get inside and see what's going on.  Trouble is we can't shrink down small enough to walk inside a silicon chip. But we can go the other way; we can build the thing big enough that we can walk inside it. Not only that we can also put LEDs on everything so we can actually SEE the data moving and the logic happening. It's going to be great.

If he believed this, he should make it out of relays, that would show the world how it really works.  This guy already did it, though, and obviously many tens of thousands of transistorized computers were built before the IC came to market.

http://www.nablaman.com/relay/

[edavid]

Did he post schematics anywhere?

I was surprised that he took the approach of building simple gate modules, because if you are doing that, why not just use IC gates?  The fun of transistor level logic design is inventing crazy minimized functions.

Also, he made the computer architecture way more complicated than necessary (e.g. autoincrement).

[EEVblog]

I've started similar but admittedly much, much less ambitious projects myself and always get to the point where I think "Well we know how to build gates using transistors what's the point in not using logic IC's? And now we can make a carry lookahead adder, get that to subtract etc why not use an MPU..."

That's the slippery slope isn't it.
Personally I'd prefer to do it with 7400 logic, just because.

[codeboy2k]

I haven't read his whole website yet to determine his reasons for his choices.. but first off, if his goal is to "get inside and see what's going on" then he doesn't need to go lower than 7400 logic.  That's good enough.

But second, since he decided to use discrete parts to build gates on PCB's, I can't believe he's using through-hole parts, like TO-92's and leaded resistors.  Unless he got a shitload of them for free, it's way cheaper these days to use smd parts (in both board size and component costs)

It would also make his 2 room beast likely fit on just 1 wall of a single room.

(PS: I won't buy any argument that through-hole is "old-school".  If that's the reason for through-hole parts, then he should be making a tube computer.  That's older-school and through-hole too)

[GK]

74XX has crap noise immunity and only a numpty would consider building with them a large scale demonstration processor (as per the construction under discussion) down to the gate (NAND/NOR/NOT) level where individual stages may be connected via several feet of wire.





 

[McBryce]

I've started similar but admittedly much, much less ambitious projects myself and always get to the point where I think "Well we know how to build gates using transistors what's the point in not using logic IC's? And now we can make a carry lookahead adder, get that to subtract etc why not use an MPU..."

That's the slippery slope isn't it.
Personally I'd prefer to do it with 7400 logic, just because.

Because... You like pulling as many amps as possible? How about 4000 Series CMOS?

McBryce.

[tautech]

I have to get my EDUC16 website started.

Please link here when you do. 

[Mechanical Menace]

The answer is in the original motivation. Did you want to find out if you meet the challenge of not using them? Well, is the answer going to be no?

The original motivation was purely educational, for myself and whoever was interested. Writing games for old basic systems got me interested in computers and then electronics, taking what I've learned and starting from the bottom up should (and has) taught me more, or at least given me a different perspective on the problem.

Hopefully the answer of if I'm up to the challenge won't be no, but I get sidetracked easily and over obsess over details like diode or transistor matrix for the ROM?, lol.

Don't get me wrong. I admire the brave attempt of actually starting. But I was looking forward to watching your progress.

And on that one I think you must be thinking of someone else. But pretty soon I should have a couple of months with nothing to do so I should get at least one DIY computer project into good enough shape to share. Should...

[John Coloccia]

I think some of you guys are completely missing the point.  If he wanted to simply build a computer, he would just simply build a computer.  The goal is to build it out of discrete components.  There's no point using SMD to reduce the size.  Have you seen his board layout?  Does it look like he's trying to reduce the size?  He's doing it like this for the hell of it, because he can.  Frankly, it's the only thing that makes this project super cool.  If he started simplifying it in any way, the end result would be a very lame, worthless hunk of junk.  Doing it the way he's doing it turns it into the electronic equivalent of a kinematic sculpture.

[nctnico]

I've started similar but admittedly much, much less ambitious projects myself and always get to the point where I think "Well we know how to build gates using transistors what's the point in not using logic IC's? And now we can make a carry lookahead adder, get that to subtract etc why not use an MPU..."

That's the slippery slope isn't it.
Personally I'd prefer to do it with 7400 logic, just because.

Why not ECL logic During an internship I got a (last) glimps of mainframes build from discrete TTL and ECL logic. Huge 5V power supplies and lots of fans. On some board they did use programmable logic (PLD) though where other boards had Z80s to offload tasks like dealing with serial ports. I vividly recall having to repair a SCSI controller with 50 (or so) programmable logic chips on it. They told me the problem with that particular board was that one of the PLDs would fail but they didn't know which one. The only solution was to pull each PLD from it's socket and compare using a PROM programmer. Fortunately for me it was the 20th or so chip that turned out to be bad...

[amyk]

I was surprised that he took the approach of building simple gate modules, because if you are doing that, why not just use IC gates?  The fun of transistor level logic design is inventing crazy minimized functions.

Agreed. What logic family is he using? From the pictures it looks like BJTs and NRTL to me, but his AND-2 and OR-2 both use 5 transistors, which doesn't remind me of anything, although the fact that he gets inverters "for free" sounds like (P)ECL?

Also, he made the computer architecture way more complicated than necessary (e.g. autoincrement).

Actually, it seems more limited than a Z80, which has fewer transistors but more instructions and registers.

If I were doing this (and I have considered it...) I'd definitely use SMD, not really for the density but for the ease of assembly - just paste/place all the parts and cook them in the reflow oven.

[edavid]

I was surprised that he took the approach of building simple gate modules, because if you are doing that, why not just use IC gates?  The fun of transistor level logic design is inventing crazy minimized functions.

Agreed. What logic family is he using? From the pictures it looks like BJTs and NRTL to me, but his AND-2 and OR-2 both use 5 transistors, which doesn't remind me of anything, although the fact that he gets inverters "for free" sounds like (P)ECL?

I think maybe NMOS, since I don't see base resistors.  Funny that he doesn't say anything about his actual circuits.

Oh, and he should work on speeding up his slow paths

[zapta]

Because... You like pulling as many amps as possible? How about 4000 Series CMOS?

McBryce.

The red LEDs on each signal suggest that low power and high speed are not design goals.

[Kostas]

I was surprised that he took the approach of building simple gate modules, because if you are doing that, why not just use IC gates?  The fun of transistor level logic design is inventing crazy minimized functions.

Agreed. What logic family is he using? From the pictures it looks like BJTs and NRTL to me, but his AND-2 and OR-2 both use 5 transistors, which doesn't remind me of anything, although the fact that he gets inverters "for free" sounds like (P)ECL?

I think maybe NMOS, since I don't see base resistors.  Funny that he doesn't say anything about his actual circuits.

Oh, and he should work on speeding up his slow paths

If you look carefully at the photograph of the frontpage, you will notice that the transistors are Fairchild's 2N7000 N-channel mosfets, so it is highly  likely that he made a NMOS implementation. The extra mosfets at the gate circuits are for the leds (one for each led), not for the gate itself, something that he mentions somewhere in there.

[edavid]

If you look carefully at the photograph of the frontpage, you will notice that the transistor's are Fairchild's 2N7000 N-channel mosfets, so it is highly  likely that he made a NMOS implementation. The extra mosfets at the gate circuits are for the leds (one for each led), not for the gate itself, something that he mentions somewhere in there.

Thanks.  Of course that raises the question of why he needs separate LED drivers... he could just use negative logic, and drive the LED cathodes from his regular gate outputs.

[Howardlong]

It'd be great to see the schematic of the gates he's using, I haven't quite figured out why the speed is so slow, although looking at the edges on a scope trace I saw I may be mistaken but it looks like an open collector/drain pullup rather than totem pole style may be implemented. I stand to be corrected of course.

Irrepective, while we're all here just talking about it and having our 2c, this guy's actually _doing_ it. For that I believe he should be commended and encouraged. There is a huge difference between the likes of you and I suggesting it should be done like this or like that, he's drawn his line in the sand and is getting on with it. Equally, I was rather taken aback by the El Reg response, it's universally positive which was encouraging.

Edit: typo due to operator error on tablet touch keyboard.

[GK]

I have to get my EDUC16 website started.

Please link here when you do. 

No

[amyk]

If you look carefully at the photograph of the frontpage, you will notice that the transistors are Fairchild's 2N7000 N-channel mosfets, so it is highly  likely that he made a NMOS implementation. The extra mosfets at the gate circuits are for the leds (one for each led), not for the gate itself, something that he mentions somewhere in there.

You have extremely good eyes!

Those are far more sensitive to ESD than BJTs, so I hope he has taken some very good ESD precautions when working on the system. Very long runs of wire, and with no protection diodes either... and on his progress page you do see mentions of dead transistors.

[JoeN]

I think some of you guys are completely missing the point.  If he wanted to simply build a computer, he would just simply build a computer.  The goal is to build it out of discrete components.  There's no point using SMD to reduce the size.  Have you seen his board layout?  Does it look like he's trying to reduce the size?  He's doing it like this for the hell of it, because he can.  Frankly, it's the only thing that makes this project super cool.  If he started simplifying it in any way, the end result would be a very lame, worthless hunk of junk.  Doing it the way he's doing it turns it into the electronic equivalent of a kinematic sculpture.

The thing that should be pointed out is that this is hardly new.  It was the standard way of achieving digital computers from the mid 1950s through the mid 1960s before ICs became available.  I'm sure examples still exist.

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

[tggzzz]

If you look carefully at the photograph of the frontpage, you will notice that the transistors are Fairchild's 2N7000 N-channel mosfets, so it is highly  likely that he made a NMOS implementation. The extra mosfets at the gate circuits are for the leds (one for each led), not for the gate itself, something that he mentions somewhere in there.

You have extremely good eyes!

Those are far more sensitive to ESD than BJTs, so I hope he has taken some very good ESD precautions when working on the system. Very long runs of wire, and with no protection diodes either... and on his progress page you do see mentions of dead transistors.

Sounds like he is trying to replicate the operational characteristics of valve computers.

[tggzzz]

The fun of transistor level logic design is inventing crazy minimized functions.

Just so.

The first computer I used was an Elliott 803, one of which is now in the National Museum of Computing at Bletchley Park. Since transistors were expensive, each gate used only ony one germanium transistor. http://theadamsons.co.uk/elliott803.htm