4 ms·

The Min-T Processor

52 points by mabynogy 10y ago

lelandbatey 10y ago
This reminds me of another educational CPU that's also in progress, the "8-bit breadboard computer" being created by Khan Academy's Ben Eater: https://www.youtube.com/playlist?list=PLowKtXNTBypGqImE405J2565dvjafglHU https://www.youtube.com/playlist?list=PLowKtXNTBypGqImE405J2... I find his diagrams and narrations extremely clear and straightforward. I've been using it as a chance to learn more about digital logic, and hobbyist electronics. For example, I've been (very slowly) learningttfdfd KiCad by translating Mr. Eaters designs into KiCAD form so they can be manufactured as dedicated boards. I finally got the first module, the clock module, correct and manufactured, and you can find the designs here: https://github.com/lelandbatey/clock-module https://github.com/lelandbatey/clock-module Here's the clock module on the printed PCB with all the components, and only one bodge wire needed (now fixed in the design): https://s16.postimg.org/gxi0plptv/2017-03-26_15.51.00.jpg https://s16.postimg.org/gxi0plptv/2017-03-26_15.51.00.jpg
- Limb 10y ago
  Very nice! I had the same idea, and had started laying out the designs he was using in KiCAD as well. If you haven't already I suggest picking up a copy of Digital Computer Electronics [0] by Malvino, which he uses as a basis for his designs. It's a great reference and lets you dive a little deeper into the design as well. [0] https://www.amazon.com/Digital-Computer-Electronics-Albert-Malvino/dp/0028005945/ref=tmm_hrd_swatch_0?_encoding=UTF8&qid=&sr= https://www.amazon.com/Digital-Computer-Electronics-Albert-M...
FullyFunctional 10y ago
Yes, this is great fun. I have a few paper designs of my own in a similar spirit and if you look around you'll find many existing projects. However the "Min-T" name to me implies a higher ambition than "just" implementing a transistor computer. True minimalism requires exacting definitions and metrics. That could be pretty interesting, but keeping memory outside the definition is in a sense cheating as [the size and speed of] random access memory was a key differentiators between early computers (ENIAC era). Essentially, this design would be facing the same constraints as _early_ computers, so I would expect the solution to look similar: micro-coded bit-serial design, Rope memory for ROM, and core memory for "ram". I'd look to the HP-1900A for inspiration.
- tonyRobinson 10y ago
  I'm experimenting with names. You'll see a write up where I minimise the number of legs, i.e. solder joints you have to make. That seemed like a very good metric. But then I found myself discussing everything in terms of the transistor count, so I'm favouring min-T right now. Passive components like resistors and didoes you can see what they do from looking at them. Transistors are much harder to work out what they are going to do. Other name suggestions are welcome.
throwaway7645 10y ago
Yes, most people can barely fathom how (well anything) works around them. The power grid is stupidly complicated (I've been studying it for 10 years) and so are modern computers and compilers, electronics, cars, aircraft...etc. I'm an engineer and am blown away by how we can work together to put a man on the moon, while individually we can do so little. I got off topic, but we REALLY need to find ways to teach tech and carry it forward.
theaustinseven 10y ago
If your going for a small processor, you should really implement a small subset of MIPS. It is really simple, and you can even compile programs from C and run them on your processor. Some Universities use this as their primary teaching tool for processor design.
- tonyRobinson 10y ago
  I liked the challenge of implementing my own instruction set. It turned out to be quite similar to a subset of ARM, so it may also be similar to a subset of MIPS (I haven't checked).
  - theaustinseven 10y ago
    Yeah, I can understand the appeal of implementing your own, but MIPS itself is very small. There are also a lot of good resources on it, and the spec fits on one page : http://www.mrc.uidaho.edu/mrc/people/jff/digital/MIPSir.html http://www.mrc.uidaho.edu/mrc/people/jff/digital/MIPSir.html
    - tonyRobinson 10y ago
      Thanks for that link. I'm sure I've looked at MIPS before and it may well have unconsciously influenced me, or it may be a case of converging evolution, but I use the three register design much like MIPS does. However, mine is much smaller, essentially only six "ALU" instructions D = S op T for op in AND, OR, XOR, ADD, SUB, ASR along with PUSH and POP. Given the simplicity of my set I can start from anywhere and cut it down, in fact I'm thinking about cutting down TCC x86_64 to what I use in order to learn how to port TCC.
theamk 10y ago
This project specifically does not try to implement memory, leaving this to Raspberry PI. In this case (1), the simplest CPU (by part count) is a 1-bit one, such as Motorola MC14500B. The application handbook http://tinymicros.com/mediawiki/images/e/ec/MC14500B_Handbook.pdf http://tinymicros.com/mediawiki/images/e/ec/MC14500B_Handboo... (pdf) explains how it works. - CPU does not contain program counter: instead, a regular binary counter counts PC from 0 to max program ROM, then wraps. You can clear OEN bit to turn write operations into no-op's. (For an advanced usage, you have JMP and RTN instructions which just set corresponding pins. It is up to you to wire something to them.) - CPU does not contain instruction decoder: instead, some bits from program ROM are wired to address bus. For example, 16-bit wide ROM will be split into 4 bit instruction and 12 bit memory address. Max amount of RAM supported depends on your ROM's width. RAM is bit-addressable. There is no indirect RAM access (unless you implement it yourself). - CPU does not contain wide registers: you get one data bit, one result bit, and two flags, for 4 bits of state total. - CPU does not contain ALU: you can do AND/OR/XOR, optionally inverted -- so just a single bit LU unit. This machine is not Turing complete at all (2), but it is still possible to implement basic arithmetic, as well as simple games such as "pong". (1) I am ignoring all other requirements here, like clear layout and educational value. (2) Strictly speaking, no real machine is Turing complete, because Turing machines require infinite memory. But this machine is super bad because it has no indirect memory access at all.
- delinka 10y ago
  "...this machine is super bad because..." I can't tell if you mean it's awesome, or terrible.
  - striking 10y ago
    My opinion is that both are true.
- tonyRobinson 10y ago
  Yes, I did consider implementing a simpler processor that ran a bit sliced ALU. I decided against it because registers are needed to see how the processor works (how it fetches and stores instructions, how the PC controls flow, how you can read in and out of memory using registers, how the ALU operates on registers). The original designs used a 1 bit ALU, then I decided that this wasn't intuitive, it would be a better learning experience and not many more transistors to have a 16 bit ALU.
jfoutz 10y ago
One of the funnest video game things i've ever done is build a little computer in minecraft, back when redstone was first introduced. just 4 bits, input as switches and output as torches. Add, subtract, multiplication, jump if zero. the 256 bits of memory took acres of space. absolutely useless excercise, aside from being a fun way to spend some time.
pankajdoharey 10y ago
I think Implementing a 6502 is a good example for an architecture implementation. Since it has already been done in numerous NES emulation projects, A lot of us have a good understanding of the CPU. Plus it is the part of a history which many kids could relate to. The video game era and genesis of personal computing era was almost entirely based off of this single CPU. It isnt small but definitely a stepping stone towards understanding a modern CPU. anything lesser makes it just a toy. For a toy, I like the one described in "Elements of Computign systems" here https://www.amazon.com/Elements-Computing-Systems-Building-Principles/dp/0262640686 https://www.amazon.com/Elements-Computing-Systems-Building-P...
- Retric 10y ago
  I would go with the Intel 4004 for historic reasons https://en.m.wikipedia.org/wiki/Intel_4004 https://en.m.wikipedia.org/wiki/Intel_4004 or other 4bit CPU for simplicity. Latency is going to force any large scale CPU into toy category, but you can still do a lot even at a few kHz with a 4 bit CPU.
  - tonyRobinson 10y ago
    8 bit and 4 bit CPUs are horrible things to teach. By being 16 bits throughout you can do many high level things in one instruction. It also makes for an efficient FORTH implementation.

52 points by mabynogy 10y ago

This reminds me of another educational CPU that's also in progress, the "8-bit breadboard computer" being created by Khan Academy's Ben Eater: https://www.youtube.com/playlist?list=PLowKtXNTBypGqImE405J2565dvjafglHU https://www.youtube.com/playlist?list=PLowKtXNTBypGqImE405J2... I find his diagrams and narrations extremely clear and straightforward. I've been using it as a chance to learn more about digital logic, and hobbyist electronics. For example, I've been (very slowly) learningttfdfd KiCad by translating Mr. Eaters designs into KiCAD form so they can be manufactured as dedicated boards. I finally got the first module, the clock module, correct and manufactured, and you can find the designs here: https://github.com/lelandbatey/clock-module https://github.com/lelandbatey/clock-module Here's the clock module on the printed PCB with all the components, and only one bodge wire needed (now fixed in the design): https://s16.postimg.org/gxi0plptv/2017-03-26_15.51.00.jpg https://s16.postimg.org/gxi0plptv/2017-03-26_15.51.00.jpg

Limb 10y ago
Very nice! I had the same idea, and had started laying out the designs he was using in KiCAD as well. If you haven't already I suggest picking up a copy of Digital Computer Electronics [0] by Malvino, which he uses as a basis for his designs. It's a great reference and lets you dive a little deeper into the design as well. [0] https://www.amazon.com/Digital-Computer-Electronics-Albert-Malvino/dp/0028005945/ref=tmm_hrd_swatch_0?_encoding=UTF8&qid=&sr= https://www.amazon.com/Digital-Computer-Electronics-Albert-M...

FullyFunctional 10y ago

Yes, this is great fun. I have a few paper designs of my own in a similar spirit and if you look around you'll find many existing projects. However the "Min-T" name to me implies a higher ambition than "just" implementing a transistor computer. True minimalism requires exacting definitions and metrics. That could be pretty interesting, but keeping memory outside the definition is in a sense cheating as [the size and speed of] random access memory was a key differentiators between early computers (ENIAC era). Essentially, this design would be facing the same constraints as _early_ computers, so I would expect the solution to look similar: micro-coded bit-serial design, Rope memory for ROM, and core memory for "ram". I'd look to the HP-1900A for inspiration.

tonyRobinson 10y ago
I'm experimenting with names. You'll see a write up where I minimise the number of legs, i.e. solder joints you have to make. That seemed like a very good metric. But then I found myself discussing everything in terms of the transistor count, so I'm favouring min-T right now. Passive components like resistors and didoes you can see what they do from looking at them. Transistors are much harder to work out what they are going to do. Other name suggestions are welcome.

throwaway7645 10y ago

Yes, most people can barely fathom how (well anything) works around them. The power grid is stupidly complicated (I've been studying it for 10 years) and so are modern computers and compilers, electronics, cars, aircraft...etc. I'm an engineer and am blown away by how we can work together to put a man on the moon, while individually we can do so little. I got off topic, but we REALLY need to find ways to teach tech and carry it forward.

theaustinseven 10y ago

If your going for a small processor, you should really implement a small subset of MIPS. It is really simple, and you can even compile programs from C and run them on your processor. Some Universities use this as their primary teaching tool for processor design.

tonyRobinson 10y ago
I liked the challenge of implementing my own instruction set. It turned out to be quite similar to a subset of ARM, so it may also be similar to a subset of MIPS (I haven't checked).
- theaustinseven 10y ago
  Yeah, I can understand the appeal of implementing your own, but MIPS itself is very small. There are also a lot of good resources on it, and the spec fits on one page : http://www.mrc.uidaho.edu/mrc/people/jff/digital/MIPSir.html http://www.mrc.uidaho.edu/mrc/people/jff/digital/MIPSir.html
  - tonyRobinson 10y ago
    Thanks for that link. I'm sure I've looked at MIPS before and it may well have unconsciously influenced me, or it may be a case of converging evolution, but I use the three register design much like MIPS does. However, mine is much smaller, essentially only six "ALU" instructions D = S op T for op in AND, OR, XOR, ADD, SUB, ASR along with PUSH and POP. Given the simplicity of my set I can start from anywhere and cut it down, in fact I'm thinking about cutting down TCC x86_64 to what I use in order to learn how to port TCC.

theamk 10y ago

This project specifically does not try to implement memory, leaving this to Raspberry PI. In this case (1), the simplest CPU (by part count) is a 1-bit one, such as Motorola MC14500B. The application handbook http://tinymicros.com/mediawiki/images/e/ec/MC14500B_Handbook.pdf http://tinymicros.com/mediawiki/images/e/ec/MC14500B_Handboo... (pdf) explains how it works. - CPU does not contain program counter: instead, a regular binary counter counts PC from 0 to max program ROM, then wraps. You can clear OEN bit to turn write operations into no-op's. (For an advanced usage, you have JMP and RTN instructions which just set corresponding pins. It is up to you to wire something to them.) - CPU does not contain instruction decoder: instead, some bits from program ROM are wired to address bus. For example, 16-bit wide ROM will be split into 4 bit instruction and 12 bit memory address. Max amount of RAM supported depends on your ROM's width. RAM is bit-addressable. There is no indirect RAM access (unless you implement it yourself). - CPU does not contain wide registers: you get one data bit, one result bit, and two flags, for 4 bits of state total. - CPU does not contain ALU: you can do AND/OR/XOR, optionally inverted -- so just a single bit LU unit. This machine is not Turing complete at all (2), but it is still possible to implement basic arithmetic, as well as simple games such as "pong". (1) I am ignoring all other requirements here, like clear layout and educational value. (2) Strictly speaking, no real machine is Turing complete, because Turing machines require infinite memory. But this machine is super bad because it has no indirect memory access at all.

delinka 10y ago
"...this machine is super bad because..." I can't tell if you mean it's awesome, or terrible.
- striking 10y ago
  My opinion is that both are true.
tonyRobinson 10y ago
Yes, I did consider implementing a simpler processor that ran a bit sliced ALU. I decided against it because registers are needed to see how the processor works (how it fetches and stores instructions, how the PC controls flow, how you can read in and out of memory using registers, how the ALU operates on registers). The original designs used a 1 bit ALU, then I decided that this wasn't intuitive, it would be a better learning experience and not many more transistors to have a 16 bit ALU.

jfoutz 10y ago

One of the funnest video game things i've ever done is build a little computer in minecraft, back when redstone was first introduced. just 4 bits, input as switches and output as torches. Add, subtract, multiplication, jump if zero. the 256 bits of memory took acres of space. absolutely useless excercise, aside from being a fun way to spend some time.

pankajdoharey 10y ago

I think Implementing a 6502 is a good example for an architecture implementation. Since it has already been done in numerous NES emulation projects, A lot of us have a good understanding of the CPU. Plus it is the part of a history which many kids could relate to. The video game era and genesis of personal computing era was almost entirely based off of this single CPU. It isnt small but definitely a stepping stone towards understanding a modern CPU. anything lesser makes it just a toy. For a toy, I like the one described in "Elements of Computign systems" here https://www.amazon.com/Elements-Computing-Systems-Building-Principles/dp/0262640686 https://www.amazon.com/Elements-Computing-Systems-Building-P...

Retric 10y ago
I would go with the Intel 4004 for historic reasons https://en.m.wikipedia.org/wiki/Intel_4004 https://en.m.wikipedia.org/wiki/Intel_4004 or other 4bit CPU for simplicity. Latency is going to force any large scale CPU into toy category, but you can still do a lot even at a few kHz with a 4 bit CPU.
- tonyRobinson 10y ago
  8 bit and 4 bit CPUs are horrible things to teach. By being 16 bits throughout you can do many high level things in one instruction. It also makes for an efficient FORTH implementation.