back to article If you want to make your own chip and aren't Microsoft rich, who do you turn to?

The likes of Google, Facebook, and Microsoft can design custom chips and have them manufactured using their billions of dollars in the bank. Smaller outfits wishing to make their own processors or microcontrollers, say, aren't as fortunate as they lack the funds and resources. Getting even a small volume of physical chips in …

  1. david 12 Silver badge

    How do they compare?

    How do they compare to other companies in the same space? Have the companies that were around 40 years ago gone out of business, or shifted to a different market?

    1. diodesign (Written by Reg staff) Silver badge

      Re: How do they compare?

      There are some others. Eg,

      MOSIS – this takes the same multiple-designs-per-wafer approach and works with TSMC, Intel, and GF. These have been around for ages.

      Europractice is another broker for access to foundries

      IC Alps often pops up

      Minimal Fab Nederland is gauging interest

      There is a choice. If there's an interest in low-volume chip making, we'll explore the area more and speak to others. I dream of having the time and budget to design and make a simple vulture chip for an article series.


      1. cookieMonster Silver badge

        Re: How do they compare?

        How about funding it here (or partially) offer a mug or something and/or an actual chip.

        I’d love to add another Reg mug to my collection

        1. Bitsminer Silver badge

          Re: How do they compare?

          There's mugs?

          1. DJO Silver badge

            Re: How do they compare?

            Yes, they buy the mugs.

  2. Pascal Monett Silver badge

    "Skywater Technology's 130nm process"

    That is not a "not advanced" technology, that is dinosaur technology these days.

    I'm wondering if it wouldn't be cheaper and more efficient to purchase a low-end Intel or AMD chip and program that. Even a 14nm general design would almost certainly be faster than a dedicated 130nm design, and more energy-efficient.

    Of course, not if you're using Windows on it . . .

    1. diodesign (Written by Reg staff) Silver badge

      Re: "Skywater Technology's 130nm process"

      130nm is OK for low-volume simple devices and microcontrollers. Academic studies, first-time tape-outs, one-off silicon, weird mixed-signal stuff, etc. No one expects it to be advanced. Depending on how much money you have, the 130nm is virtually free (or free if qualifying with Google), so you get what you pay for.

      Skywater does 90-350nm (and seems to be trying out 65nm) and if you need something else, like 65nm and below, there are other foundries.

      BTW one of the reasons why you want a high transistor count (and therefore a dense node) is to fit a decent amount of cache on the die. If you don't need a lot of cache, and you're not doing an SoC with a lot of complex things on it, well, why do you need a small node?

      But hey, don't let me stop you. If you want something super dense, knock yourself out. Just cough up the six or seven figures rather than ten large.

      For instance, take SiFive: their first chips were in the 16-14nm range, and it's had hundreds of millions of dollars in funding, and sales on top of that.


    2. Justthefacts Silver badge

      Re: "Skywater Technology's 130nm process"

      Yes. Yes it would. This is a teaching tool, which there is absolutely a place for.

      As a practical device, at the low end, you can buy an off-the-shelf microcontroller that will do anything you want as a board controller. And at the higher end (“does it run Linux”), you won’t beat an off-the-shelf ARM or AMD/Intel for cost, performance and power.

      Just possibly, the penny is starting to drop: none of these companies are even *considering* 14nm, which as you say is still a legacy node. And that’s because making something cost and power competitive on 14nm will cost *whoever does it* hundreds of millions. The “big companies” aren’t cheating you. That’s just how complex the job really is. Having an open-source Verilog to slap down doesn’t even get you the first 1% of the way there.

      1. Yet Another Anonymous coward Silver badge

        Re: "Skywater Technology's 130nm process"

        These custom runs are actually more useful at 'obsolete' process sizes.

        A lot of them are power chips, some RF with novel antennae design - which need big gates.

        Even big fabs (like TSMC) are reasonably cheap/accomodating for getting prototype quantities made on these processes. And the design tools are cheap(ish) and well known

        If you need custom 5nm chip then A) you are going to need multiple years of process development work with TSMC before you can even think of taping out a sample, and B) you can probably just wait for a regular fast CPU/SOC to come out and do it in software.

  3. Tom 7

    Upfront costs?

    A few years ago I looked into knocking up some HV FET stuff for connection to a central controller for a modular Inverter/charger/MPPT and was told it would cost me $40,000 just for the spice parameters to simulate the process I was trying to get access to, without which I couldn't even be sure the process would do what I want!

    I'm just completing some designs for a wind generator that should pay for itself in a couple of years (no planning permission required!) but thats for 12/24V output and integrating it into the grid/house system would require A) putting an OCR on the useless smart meter to judge prices and consumption, B)an inverter that would double the cost of the system despite only needing to cost about £100 more than the bloody legally required transformer interface, C) a whole other collection of over-expensive bits and pieces to get in the way of using your own power and sharing the excess with the grid (or not should that stop working and prevent you using your own power.

  4. John Savard

    Speaking of Nodes

    It's true that for some purposes, 130 nm is a disappointment. But if 65 nm were ever to become available to ordinary mortals at an affordable price, that would be enough, since Intel's Core 2 processors made on that node ran at 3 GHz, just about as fast as current microprocessors - and they were Core 2 processors, and so comparable to current ones, as opposed to the Pentium 4, which attained high clock rates by having very short pipeline stages.

    1. mithro

      Re: Speaking of Nodes

      The best way to getting a fully open source 65nm or 45nm process technology is to help make the 130nm programs (and other future programs) successful. If the community can show the demand for open source process technologies is there and it is enabling an explosion in new designs and customers for the foundry then the argument will be easy.

      There is a document at which aims to provides some inspiration for what can be done with the currently available technology.

      1. Justthefacts Silver badge

        Re: Speaking of Nodes

        I still don’t understand what the end goal is? What are these “new designs” that aren’t just easier to implement in software?

        And not just easier, more power-efficient. Never mind 130nm, you are going to get much better power and performance doing anything that is “processing” in software on a 7nm general purpose CPU, than on full-custom silicon at 45nm. That’s why RF and power-engineering ASICS are far more useful to the “DIY full custom” community, and that’s just as good on 130nm.

        And by the way, something that people always forget: you never get a “fair” comparison between 7nm off-the-shelf and 7nm custom-built. It takes minimum 3 years to make an ASIC including volume rampup. So even if you started a full-custom development today on 7nm….by the time it gets to market, it is competing head-to-head with an off-the-shelf 2nm device.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like