DTL Stack Computer

vidak

I was told something by my best mate's dad on the weekend. It was that, in the 60s and 70s, the entire province I lived in had one computer for the entire government. I tell no lies.

It had 64K of core memory. This is a massive computer, but it shows how important permacomputing is.

So I am going to construct a moderately powerful diode-transistor logic computer out of discrete components. I will the attempt to network it, and produce others.

i am having issues adding images. i wanted to upload all my screenshots and PDFs here.

vidak

NB: http://www.6502.org/users/dieter/mt15a/mt15a_3.htm

vidak

http://www.6502.org/users/dieter/mt15/mt15_sch.htm

stman

As you (should have) understood from all what I told you regarding FPGA, they do contain some backdoors too. We know what kind of backdoors they are, but this is the only thing we know. There are no details accessible to simple citizens like us. We know how to mitigate them on the assumptions we do regarding what they are, but there is no garantee that something more clever has been inserted, making all the mitigation technics I mentioned to you useless. Still, there is no magic, there is only science here. Cryptography is just a way of playing with time as your ally. Time is a very reliable ally as nobody can buy time. Time is time and this is all. Playing with time is safe, if you know exactly what you're doing. This is usually the case, but side channels can make the game harder. Indeed time is a good ally as long as side channel cannot buy time.

Discrete logic :
Discrete logic gates (74xxx) is a good option as long as you can reasonably ensure that they are not backdoored. Normally, nobody backdoors fundamental logic gates. But.... this is just normally. We don't live in a normal world any longer. The worst scenario I see here is people breaking into your home and changing a few 74xxx gates on your PCB for backdoored versions, in strategic places. This is a highly paranoid scenario. In this scenario, what is important is to be sure about your supply chain for those 74xxx components. And here there is a problem. You can never be sure. Even if you are, this is not solving the problem of people breaking into your home secretely to unsolver a few 74xxx components on your PCB and replace them with backdoored ones. But let's takcle problems one at a time. The supply chain issue can be mitigated by getting your 74xxx components from old used machines, unsoldering them for reuse them on your PCB. Still, this only solves the supply chain issue, it does not solve the problem of people breaking into your home to unsolder a few 74xxx on your PCB and replace them with backdoored ones.

So now the last problem to solve, is how to ensure such thing can happen. How to mitigate NSA TAO masters. And here the game is really harder. The first thing that comes to my mind, is to personalize every 74xxx on your PCB so that those wanting to replace them with fake backdoored ones can't do it without being noticed. But what does personalization means ? It means finding a way to make those packages unique, and be able to read/control something on the package of each IC to ensure it is really the one you soldered. To do this, many cryptographers would tell you to alter the package in a random way to that is become unique and not easily reproductible. There are many ways to customize IC packages so that they become unique. You can choose to put some glue on them, and through some glass powder on the glue while it is still not dry. Doing so, you create a fingerprint that is hardly reproductible. Doing so you prevent NSA TAO or equivalent to swap IC on your PCB. Still, you have to have an accurate way of scanning your PCB (taking like microscope photos of it) so that you can read easily and accurately the fingerprint you created with the glue. Proceeding this way, you can reasonabily be sure that if the fingerprint is altered, it means somebody tried to tamper with some of your IC's on your PCB.

Now, if you go to discrete transistors, it is exactly the same issue. You will have to find a reliable way to create a fingerprint that, if altered, would mean somebody tried or tampered with your transistors.

All this shit, and it's quite a load of work, if the consequence that we still don't have free integrated circuits.

The technics using CD/DVD OPU (Optical Pickup Units) to scan several parts (IC's) or whatever on your PCB is a reasonbily secure approach to check a self made fingerprint authenticity.

But hey, have you read the thesis on OPU's used as low cost microscopes ? It's cheap, but it's still some work to have all this fully operational.

The situation is to be considered almost desperate dear friends. At least one thing is fucking sure, it's that there is only the supply chain issue to solve, to acquire those components (74xxx discrete logic or transistors), the second problem is to fight any unwanted swapping (By NSA TAO or equivalent) of those unbackdoored chips on a PCB of a device you would have at your place.

It's not that solving all this is impossible, it's just work. Hard work to make it as simple, cheap, and reliable as possible.

I never said it was impossible, I just say it's hard work.

vidak

i think with transistors, surely the only thing you'd need is a cheap lab microscope?

i am going to be building it out of discrete transistors.

stman

vidak My last post was too long and too paranoid.

Building your processor with discrete transistor is clearly the safest, but the PCB can be huge.
I think building around 74xxx logic or similar (more recent gates, comutting faster) is the bet compromize between safety and integration.

I was too paranoid with my scenario where evil entities would break into your home and swap a few strategic 74xxx gates on your PCB by backdoored ones. Still, as I said too, it is easy to ensure that nobody will unsolder them to swap a few of them, by creating a unique and not reproductible fingerprint on your whole PCB : Using some transparent glue and using some debris / small pieces of glass, almost like sand grains, and gluying them on the PCB makes your PCB protected against tempering and swapping of IC's, with a visually easy to check fingerprint form of the glue + glass sand glued on the whole PCB like a layer of paint. All you need to do is to keep a safe original photo of the fingerprint once it is created, and then, it's easy to check that nothing changed.

Regarding the supply chain attack when you acquire those logic gates, it's easy to go to a few random stores and by them, to ensure they are not backdoored ones. This risk is high only if you buy them all at once online on a website.

Still, it is always interesting to know how far agencies teams like NSA TAO can go.

Personnaly, I am going to design and construct me a TCP/IP full stack that can manage a limited amount of sockets simultaneously made out of 74xxx or more recent elementary discrete logic gates. And I am not going to use any tool to do this, I am going to create the schematic the old way, manually. I will just prototype these FSM on FPGA first to ensure everything is fine, and then I will do the translation into 74xxx or equivalent logic gates manually.

All what is important here for easy peer review, is to maintain good FSM's diagrams of the design.

Tools are necessary only when designs start to be huge and very complex and when you want some optimizations of several kinds, but for stuff like a small TCP/IP stack, I think it is way safer to do everything manually. No VHDL, no autorouter.

We tend sometimes to forget that in the past we were doing everyhting manually, and that is was working perfectly, with no bugs, and that it was not taking more time than with using design automation tools. The advantage of doing everything manually is that it is straightforward and super safe, you don't have to assume anything regarding tools integrity as you don't use them.

stman

stman Look at this table :

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

There is a new TTL super fast (Giga hertz range) logic family introduced in 2004, the 74Gxxx family. I was not aware of it. It's a long time I did not had a look at such comparison table : http://www.potatosemi.com/potatosemiweb/74Logic.html

Let me have a look if those 74G family gates are easily available, but they seem perfect for a TCP/IP stack the way I want to do it. There is not the whole family available, but there are sufficient ones to play with. And not a whole design may need them, only a part of it where clock speeds are very fast.

For less speed critical logic, the 74AC / 74ACT (125 MHz) and 74HC / 74HCT (50 Mhz) is fulfilling most needs : What is important here is the possibility to have voltage compatibility (3.3V) between the G / AC /ACT / HC /HCT families.

For a TCP/IP stack with self made PHY operating for 10 MBPS ethernet, the HC/HCT should work, but it's the limit of these families. With AC/ACT we are sure it will work perfectly. For 100 MBPS ethernet, building your own PHY become more complicated, but the AC/ACT should still be ok to handle the TCP/IP stack, but it's the limit of such family.

Dunno if I told you, but there are some Russians crypto-anarchist who did a self made PHY for ethernet 10 MBPS.

If you want to create your own 100 MBPS ethernet PHY, you will need a super fast ADC/DAC. The encoding on the lines are more complex, but it's not impossible to implement them. For 100 MBPS ethernet, sampling of the signal shall be done at minimum 250 MHz for confort, and 500 MHz to ensure good accuracy, but I have to re-read the relevant standard to ensure it. It is clear that creating your own PHY is hard work, but once it is done, it brings a really high level of security, as we fear commercial PHY than handle 10/100/1000 MBPS ethernet are backdoored. Such backdoors could include secret tagging of all datagrams with a unique serial number, allowing further anonymity breaking. We have no proof such backdoor exist, we only assume it does because it would be fucking efficient. It's typicaly the kind of low level backdoors NSA would master.

Nobody ever took the time to reverse engineer those commercial PHYs, it's a lot of work, only a universitarian guy could spend such time doing such work, so we assume they are backdoored because backdooring them would be highly strategical to break anonymity.

@theruran used to ask me to have a look at the standards defining those PHY and I had found many "suspect" things that could lead to easy backdoor creation for secretely tagging all data with serial numbers. This is particularily true for 100 MBPS and higher speeds. But it's much harder to do such tagging with 10 MBPS standard, there are way much less (If not to say "almost no") opportunities to create and use side channels on 10 MBPS, while there are a lot for 100 MBPS and higher speed rates.

To summerize what I have just said, the 10 MBPS ethernet signal is simply encoded and require no "sampling with an ADC" to decode it, while from 100 MBPS and above, it is mandatory to use an fast ADC sampler on the signal, which makes the whole design much more complex : Creating a fast ADC from discrete logic and transistor is possible, but it's work. Using a commercial ADC / DAC to do the sampling and generation of the signal wave for 100 MBPS and above ethernet speeds to create you own PHY seems really the best compromize.

Complexity of Ethernet PHY rise a lot from 100 MBPS included and above, while it is still reasonable and accessible without too much work for 10 MBPS speed.

The work of the Russian Crypto-Anarchist who did his own PHY for 10 MBPS ethernet with an Altera FPGA :
https://hackaday.com/2016/01/02/fpga-to-ethernet-direct/
http://we.easyelectronics.ru/plis/softovyy-phy-dlya-ethernet-10base-t.html
https://github.com/iliasam/10base_t_software_PHY

vidak

I have revised my idea somewhat (!!!)

I think we should target hardware that we know we can trust, that already exists.

I think stman's suggestion that we target either the 6502 or the MC68K is a good idea.

I am going to begin attempting to write a small Lisp text editor for, say, the 68K. I think that is enough power. I know the 6502 very well, but I think we should not forbid ourselves a little bit of luxury xD

stman

The advantage if we target 68k, at least for our small group, is that I really master 68k assembly language as much as I master the french language, with more than 1.2 millions lines of code coded so far in my life, this means I can rapidly develop anything in assembly language.

vidak

nice.

i would very much like to learn from you.

what do we need in order to get a full 68k micro-computer assembled, hardware-wise?

i am most importantly thinking about the display controller

theruran

Believe it or not, I have written real 68HC12 assembly, which I suppose is like the 6809. Maybe you all can enlighten me as to its relation to the 6502 or 68000.

Chris

[unknown]
The 68k is the 16 bit successor to the 6800.

The 6501 was pin compatible with the 6800. The 6502 wasn’t pin compatible, but it worked with the same support chips. Neither was instruction set compatible. Motorola was able to kill off the 6501 with a lawsuit, but not the 6502

The 8502 person on Masto is using a W65C265S, a 16 bit version of the 6502 with a 24 bit address space - 16 MB. I just started following their project. It looks interesting and fun

theruran

Chris

I guess their similarities and differences become apparent and natural as you learn more assembly languages, just like learning many programming languages of different paradigms.

vidak

hey everyone.

back here again!

i have a lot to update you on, please wait a little while…