In 2018 I made the first homemade integrated circuits in my garage fab. I was a senior in high school when I made the Z1 amplifier, and now I'm a senior in college so there are some long overdue improvements to the amateur silicon process.
The Z1 had 6 transistors and was a great test chip to develop all the processes and equipment. The Z2 has 100 transistors on a 10µm polysilicon gate process – same technology as Intel's first processor. My chip is a simple 10×10 array of transistors to test, characterize, and tweak the process but this is a huge step closer to more advanced DIY computer chips. The Intel 4004 has 2,200 transistors and I've now made 1,200 on the same piece of silicon.Previously, I made chips with a metal gate process. The aluminum gate has a large work function difference with the silicon channel beneath it which results in a high threshold voltage (10V). I used these metal gate transistors in a few fun projects like a guitar distortion pedal and a ring oscillator LED blinker but both of these required one or two 9V batteries to run the circuit due to high Vth. By switching to a polysilicon gate process, I get a ton of performance benefits (self aligned gate means lower overlap capacitances) including a much lower Vth which makes these chips compatible with 2.5V and 3.3V logic levels. The new FETs have excellent characteristics:
[...] I was particularly surprised by the super low leakage current. This value goes up about 10x in ambient room lighting.
Now we know that it's possible to make really good transistors with impure chemicals, no cleanroom, and homemade equipment. Of course, yield and process repeatability are diminished. I'll do more testing to collect data on the statistics and variability of FET properties but it's looking good!
Linked story has pictures as well as many more details on the chip's construction.
(Score: 5, Interesting) by Rich on Tuesday August 17 2021, @02:05PM (14 children)
The YouTube algorithm already put that in my feed a few days ago. Hats off to the guy. This is beyond 1337. Although "at Home" here would be valid only for very small subsets of "Home", where "Home" is stuffed with technology that is more advanced than that seen at '70s semi fabs.
His maskless technology is great, how he modded the DLP projector (UV filter out, adapted optics, software compensation for brightness). The results with their super clean feature edges scream "excellent yield". Brought into a production process, it would allow for easy creation of many one-off ICs with a single wafer start. I could easily imagine a Shenzhen shop offering such stuff at hobbyist prices (like, say, $20 for 10 PDIP-16 ICs, or a bit extra for uncapped ones to probe in a ceramic carrier). I wonder if mask, or better, exposure images could be created with an unmodified KiCAD if you had the exposure layers in footprints for associated schematic elements.
Quick calculation says that with a 1080p projector, square die, 4 pixels per structure size, 270x270 elements could be achieved, which is in the ballpark of an original NMOS 68000. (And I think he can only do N-channel, so far).
(Score: 2) by c0lo on Tuesday August 17 2021, @02:33PM
http://sam.zeloof.xyz/category/electron-microscope/ [zeloof.xyz]
- Jeol JSM-6300 SEM
- ISI Super III A
Even second hand, these two will get easy in the $100k-$150k ballpark.
https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by looorg on Tuesday August 17 2021, @02:39PM (12 children)
It's very interesting. I wonder if this is the next area of "homemade". There are already places like PCBWay (there are many others, this one just comes to mind) that lets you print your own pcb:s etc, they later added other things. I do wonder if something like make your own IC will be their next thing. If he can do 70's stuff now I'm sure they would be able to do similar things and possibly the next decade without to many problems. I wonder how mad Intel/Motorola etc will be when people start to make their own early "retro" circuits.
(Score: 5, Insightful) by JoeMerchant on Tuesday August 17 2021, @04:03PM (11 children)
Something that always impressed me about the 6502-6811 tech was that they continued to produce it on the old fabs, with the big old feature sizes and slow clock rates.
4004 is a long way from 6502, but it's a step. I always thought that the 6502 based computers could "scale" to do most text based office tasks as well as anyone needs them to. Certainly not WYSIWYG graphic editing or videos, but character based operations, like I'm doing now, like I do 90+% of the time I'm using a computer still today...
🌻🌻🌻 [google.com]
(Score: 4, Interesting) by looorg on Tuesday August 17 2021, @04:49PM (10 children)
I see all these as steps in that direction, perhaps I'm just hoping for to much. But yes I would love them to one day be able to recreate the 6502, 6510, 6581 etc. In some regard the only way to get them now is finding older machines and grab them from them but that will only work for so long, some of them are already reaching ridiculous prices. While always an option I'm not really a big fan of the FPGA route either, it just seems like such overkill and somewhat counterproductive.
While I believe that a lot of such office work could indeed be done on a simpler machine I doubt we'll be able to convince the masses of such things. It will be hard once they have now during decades been trained that you need a GHz cpu/ram to even look at your email. But if it was just text processing of some kind then yes you don't need the fancy graphics etc. I guess the only issue I would have most of the time would be a the lack of RAM, the number-crunching can always be offloaded or run at later times.
(Score: 3, Interesting) by Rich on Tuesday August 17 2021, @05:55PM (9 children)
The 6581 is particularly desirable, huh? But that's the most difficult, because the filter relies on the behaviour of the FETs from the MOS Technology process as variable resistors.
I see two uses for this retro stuff. One is retro fun and historical preservation, pretty much the same why wealthy people had the FW 190 and the Me 262 rebuilt from scratch. The other is hedging for semi-apocalyptic conditions. I guess the airplane analogy would be an open-source project to re-build the DC-3.
I think that, for CPUs, something new between the 6809 and the 68000 would be of interest, with all the hindsight gained. (But note that the Z80 is still in production!)
(Score: 5, Interesting) by JoeMerchant on Tuesday August 17 2021, @06:11PM (1 child)
Seems like a 6502 level fab would be much less costly to develop and run today, maybe not a $1000 widget like a 3D printer, but hopefully less than $100K for low volumes. Also hopefully less environmentally demanding / sensitive to vibration, dust contamination, etc.
So, you've got an 8-bit chip fab rolling for $100K - what do you do with the output? My first instinct would be to find some niche/twist on the Arduino/Pi Pico market. I would also hope that you could possibly modernize the designs somewhat with onboard RAM/Flash/basic peripherals to make them less demanding for hobbyist use, but now you're asking the fab to be better than the ones that were running in the 1980s and I guess the cost is going up.
Fun fact: around 20 years ago, Caterpillar was still controlling their big toys: excavators, graders, backhoes, etc. using 6811s. Not sure what they're doing today, they certainly don't need much more power than that - the big iron can only move so fast.
🌻🌻🌻 [google.com]
(Score: 3, Interesting) by FatPhil on Tuesday August 17 2021, @07:07PM
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 3, Interesting) by looorg on Tuesday August 17 2021, @06:16PM (1 child)
Yes, I would say the 6581 is particularly desired. For various reasons as noted it's a pain to make. But the second hand price for a SID (or 6581) is now around $80-100 each and then you might not even get a working one. You can buy an ARMSID, I think it's probably the best reproduction (or whatever we should call it) at the moment for about $30. It's very good, I doubt most people could tell them apart.
While a 68k more be quite interesting and, possibly more, useful they are not really problem to get a hold of as far as I can recall at the moment. As far as I know they are not really in production anymore, there are variants still around but they are not identical and you couldn't just swap one for the other. But they are a quite common cpu so it shouldn't be hard to find. What is hard to find tho are the later members of the 68k line such as the 060, the 020/030/040 are probably quite problematic to find also but I have not really looked much for them. Part of the problem there is that the PPC replaced the 68k and it sort of just went away. It's now firmly in the retro section I would say. Also it seems a lot of products (non-computers) that used them just swapped over to various ARM-cpus as far as I can tell. So they solved their shortage or issue that way.
(Score: 2) by Acabatag on Wednesday August 18 2021, @02:13AM
An interesting variant on the 68K line is the 68328 [wikipedia.org] that was used in the Palm Pilot. I'm not sure if the 68K version of the Dragonball is still in production anywhere, but it's an interesting and viable 68K variant.
(Score: 2) by looorg on Tuesday August 17 2021, @07:00PM (4 children)
One thing thou I'm not sure how many of them would be needed, and in that regard viability of the production. It's not like it's a common chip and there are somewhat limited uses for it (mostly various commodore 8bit, some audio cards for other machines and some standalone music equipment). The market could probably be satisfied quite quickly if there was actual production. Which is perhaps why I'm more hoping that production somehow becomes possible and something like PCBWay (or similar) finds a way to make your own chips and you could make smaller runs etc, that said I know it's not viable to ask them to make like 1 or 10 of these quite probably. So someone would have to bulk order, or you could have a system of pre-order and once a sufficient amount of orders are in they do a couple of runs and then back to waiting again.
(Score: 4, Interesting) by Rich on Tuesday August 17 2021, @07:28PM (3 children)
Exactly not. The maskless exposure process does not care if there are different chips on a wafer (as long as they fit into the slicing/dicing pattern). You could really just make, say 5, without any extra cost. The wafer then gets diced and by means of an automated production pipeline fed to bonding. The bonding coordinates for each chip come from the same order database as the exposure images; again the machine does not care where to bond the 16 wires.
The major problem I'd see with the whole offer is that the offered processes just wouldn't offer what you specifically need and what got you to want custom silicon in the first place, be it switching times, that extra metal layer, power transistors, high voltage capability, and so on.
Also, the pessimist in me fears that with such an offer, less-than-multinational corporations might attempt to make cloning (and repairs!) harder. (But on the other hand, custom obfuscation silicon made with such an ancient process would be easy to "photocopy" (cf. the 6502 mask reconstructions)).
(Score: 2) by looorg on Tuesday August 17 2021, @08:20PM (2 children)
Interesting. I did not know that you could do that. I was apparently under the wrong impression that you made the same one in every (or per) batch.
(Score: 4, Interesting) by Rich on Tuesday August 17 2021, @09:02PM (1 child)
The guy is not only re-enacting the 70s, but really at the frontier of research. I found this article
https://en.wikipedia.org/wiki/Maskless_lithography [wikipedia.org]
which notes as last of a long list "DLP technology can also be used for maskless lithography.[8] ", with a link leading to a single Japanese company. This is definitely a mostly unexplored field. All research going into low nanometer processes, but he might tap into a "long tail" market there, where custom is needed at low quantities, and 2 micron are good enough (mixed signal?!). I also hat the thought that, because he can photographically align his stepper, he can use multiple expositions per die (say 2x2), which, in conjunction with a 4K projector would get his structure plane into the order of 1000x1000. This is already serious stuff (well into AVR territory). But I guess before going for scale, he'd want to tune his process and get good parameters for both NMOS AND PMOS.
I wonder if people have considered throwing venture capital at him, or just buying him away, so he doesn't disturb larger players.
(Score: 3, Interesting) by JoeMerchant on Wednesday August 18 2021, @12:22AM
As soon as a larger player feels threatened, he will be retired. $30m buys a nice private island, and is a trivial sum to the major players.
🌻🌻🌻 [google.com]
(Score: 2) by Opportunist on Tuesday August 17 2021, @02:47PM (9 children)
I don't want to lie, this is super impressive, but what kind of money would someone have to drop to do this "at home"? Some of the things on this page don't exactly look cheap.
(Score: 3, Informative) by Anonymous Coward on Tuesday August 17 2021, @03:14PM (8 children)
He did this in a literal garage that his parents set aside for his experimentation. And most of his equipment was bought secondhand from ebay, local 'junk shops' etc as opportunistic buys. So the cost to recreate his lab could vary widely, and these specific parts certainly might if they were common enough to be popular. The high flow pump (was it cryonic?) he was using for some part of the fabbing sounded like the most exotic part to get, but if you had some good machinist friends even some of these parts could be produced, albeit not cheaply, if you had the design in CAD or dimensions for someone to hand machine them.
That said, I seem to remember some key details missing last time I read up on it, so replicating his fab might be beyond the average person without independent research. This does however have the potential to be a major gamechanger if someone can document and streamline his process without coming to rely on expensive equipment, since these processes were the last ones you could use without getting into exotic, expensive, or toxic materials. That said, gate leakage was high so process shrinks and overall performance will be low by modern standards. But if you set your sights at ISA/PCI bus speeds and say SDRAM no faster than 33 mhz (with custom charge timings obviously) you could probably build some non-trivial systems for the retrocomputing scene, or soon enough, your custom encryption processor/storage controller for keeping your data secure in the 'Brave New World' that our government (if you're American) is seeking to force us and probably the world into. (See: Clipper Chip v2.0 in the form of backdoors, likely by those management processors and fTPMs everyone is pushing now.)
(Score: 2, Interesting) by Acabatag on Tuesday August 17 2021, @05:59PM (7 children)
If we wanted to develop our own 'security model' infrastructure, building on NOS (new old stock) and salvage parts is probably more realistic. I repair and often end up scrapping out hardware with 68HC11/8051/8048 processors on them. I've desoldered and reused IC's since back when the Z80 (and inferior parts like the 6502) was current.
(Score: 4, Touché) by FatPhil on Tuesday August 17 2021, @07:10PM (6 children)
That's fighting talk!
(And I say that as someone who had 3 different Z80-based machines, and everything I've read about the 6502 has told me I'd have had more fun programming that instead.)
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 1) by Acabatag on Tuesday August 17 2021, @11:27PM
I wish I had more experirence programming the 6802 and not it's knockoffs.
(Score: 2) by Rich on Wednesday August 18 2021, @09:18AM (4 children)
Oh, that last century...
Mac vs Windows? Same shit, or we wouldn't have the discussion here
Ford vs Chevy? They still make cars? Meh.
Blonde vs Brunette? Sigh. As long as they're not pink.
ps But for 6502 and Z-80, I've got to say that, even coming from the 6502 side, the Z80 wins hands down for native code. Chuck Peddle made a mis-judgement by assuming external memory would be as good as registers with the zero-page mode, and the absence of any 16-bit register coupling makes native code that deals with real-life quantities too big. Woz must have sensed that early on, when he wrote Sweet16.
(Score: 2) by FatPhil on Wednesday August 18 2021, @12:33PM (3 children)
A 16-bit add's just an 8-bit add followed by an 8-bit adc if you need it. But 8 bits is the perfect size for the number of remaining lives, or your character's XY coordinates, who needs 16 bits?!?!?
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by Rich on Wednesday August 18 2021, @01:56PM (2 children)
Don't forget that a Z80 on equal process runs at twice the clock. Clock for clock, the 6502 is usually almost twice as fast as the Z80 for any given problem, but given its half clock, it loses out. Another example would be a "memcpy", where the 6502's optimum (but limited) "... DEX BPL loop" takes 13 cycles, where one Z80 generic LDIR step takes 21. And to get there with the 6502, you have to jump through hoops.
To put it in a general way: The Z80 can adapt to your task, but for almost the same performance, you have to adapt your task to the 6502.
It does not help the 6502 that it effectively has to pull microcode for common operations from the system bus both from a bus load viewpoint and from code size. Turbo Pascal on a Z80 card was able to create completely reasonable programs, while the native Kyan Pascal for the 6502 generated such bloated binaries that it was effectively unusable (and its ISO string handling didn't help either). But as soon as the 6502 gets an intermediate, more dense, code layer (e.g. Woz' Sweet16 or UCSD P-Code), it can keep up.
Things get worse for the 6502 once the clock scales, because the memory can't keep up. It might stand a chance when it has fast SRAM for the zero page, and possibly the stack, but I think that's very rare. (I don't know what the ZIP Chip's ASIC did, but I guess that was hardwired)
Anyway, the true winner of the 8-bit wars should be the 6809. :)
(Score: 2) by FatPhil on Wednesday August 18 2021, @02:32PM (1 child)
The 6809 was of course going in the exact opposite direction from the 6502, it was an enhancement of the 6800 rather than a pared back version. So it's hardly surprising that it has more brute force. You paid for it in silicon though.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 3, Interesting) by Rich on Wednesday August 18 2021, @04:24PM
And while the 6502 used half of the die space, which helped it to get to a design-winning price point, we shouldn't forget that the Z80 had a much more elaborate bus system that included automatic DRAM refresh (that only uses 7 bits: one of its biggest flaws). Z-80 "pro" machines could easily be built with banks of 4116 DRAM, or even the DMA chip for massive throughput. It took a Woz to create refresh/video logic around the 6502, even Commodore themselves still used static RAM in the 1980 VIC 20.
In hindsight, we're so much smarter. The 6502 could have been made even simpler by dropping the decimal logic and the (ZP,x) addressing mode, no one would miss that, while the Z80 could have used a better refresh counter. Maybe even one that would have been suitable for driving video with minimal external glue. These guys were moving fast, and they didn't have the vision how their designs would last into a time where CPUs become much faster than RAM, or there would be denser memory than banks of 4116. Being barely good enough, but being on sale was make or break then. They would have done a mask revision to fix an ALU bug on a kitchen table in the time I wrote this lame armchair historian post. ;)
(Score: 1, Interesting) by Anonymous Coward on Tuesday August 17 2021, @03:16PM (2 children)
Teenagers have been doing this since June 1970 at least.. http://www.1010.co.uk/materials.html [1010.co.uk] - from that month's issue of Scientific American.
(Score: 4, Interesting) by Tork on Tuesday August 17 2021, @04:26PM (1 child)
I don't get the negativity, sorry.
🏳️🌈 Proud Ally 🏳️🌈
(Score: 3, Interesting) by JoeMerchant on Tuesday August 17 2021, @06:14PM
Look up: Crystal radio. They are most commonly used to receive AM, used to be a very common hobby project.
Around 1999 I assembled an FM transmitter kit using through hole components on a PCB about 4"x4" - it worked pretty well, not as well as the bluetooth-FM converters you can get in a cigarette plug that also offers USB charger outlets these days for $10 from China (I think my transmitter kit cost $99.)
🌻🌻🌻 [google.com]
(Score: 1, Interesting) by Anonymous Coward on Tuesday August 17 2021, @06:43PM (8 children)
> The Intel 4004 has 2,200 transistors and I've now made 1,200 on the same piece of silicon.
I estimate that Warren Toomey's "Crazy Small CPU" https://minnie.tuhs.org/Programs/CrazySmallCPU/ [tuhs.org] would be about 600 transistors (with the RAM and ROM still as external devices) ...so doing a CPU even at this stage could be doable.
(Score: 4, Interesting) by Dr Spin on Tuesday August 17 2021, @06:59PM (7 children)
A PDP8 has about 1,000 transistors, and probably has an order of magnitude more software than any of the other suggestions,
and das blinken lights.
They are really handy for Arduino type applications, and far easier to program in assembler than writing C++ for big iron.
It is quite hard to reduce your instruction set below 8 opcodes!
Warning: Opening your mouth may invalidate your brain!
(Score: 2) by DannyB on Tuesday August 17 2021, @07:53PM (2 children)
What if we could have only a single opcode, let's say OP, and then the arguments to it indicated whether it was a load, store, ALU op, etc.
Why is it so difficult to break a heroine addiction?
(Score: 1, Touché) by Anonymous Coward on Tuesday August 17 2021, @09:07PM
Save a bit and remove the meaningless op code.
(Score: 0) by Anonymous Coward on Wednesday August 18 2021, @03:25PM
OISC/One Instruction Set Computer. SUBLEQ is an example.
Really at the level of a Turing Machine, but I got as far as calculating factorials with mine: https://www.youtube.com/watch?v=CAtPdP_RIr4 [youtube.com]
(Score: 1) by Acabatag on Wednesday August 18 2021, @02:25AM (2 children)
The Intersil/Harris 6100 [wikipedia.org] is a single-chip microprocessor version of the PDP-8 and just looking at the die photo on the Wikipedia page, it has far more than 600 transistors. I at one point owned a PDP-8/e; they have multiple boards completely filled with 74xx era TTL gates, and the transistor count would be far greater than 600.
I have 3 or 4 full tubes of NOS Harris* 6100 processors; a retirement project I hope to launch eventually will be a 'toy' machine project that uses them. They're fascinating parts for a toy project, because they're implemented in static CMOS so can be clocked down to DC and stepped with a pushbutton if you want. Plus, who doesn't want a 12 bit machine to mess around with in Octal?
(*Harris produced the second-source 6100 part)
(Score: 2) by FatPhil on Wednesday August 18 2021, @01:24PM
Loads, I lost count at 3777.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by Dr Spin on Wednesday August 18 2021, @09:41PM
The 6100 did a lot more than a strait eight. I believe it has a complete on-board ROM for a monitor
and boot stuff.
And these days, it is easy to get software that would optimise the logic. I would not want to clone the
PNP discrete logic circuits for any number of reasons.
I agree the 8E is the one to clone, but the decision as to where to trade speed for simplicity is
wide open. I assume that speed is a low priority for most potential users today, but was not when the
6100 was born. A physical open collector I/O bus would be very useful.
Automating pet feeders does not need a whole lot of MHz.
Warning: Opening your mouth may invalidate your brain!
(Score: 2) by epitaxial on Thursday August 19 2021, @02:29AM
Even more impressive is with OS8 you could have multiple users on a PDP/8. A machine without a single integrated circuit.