What Are People Actually Making on Shuttle Runs?
We counted 3,625 designs across 19 shuttles over three years
Open-source design tools and shared wafer space have brought something remarkable within reach: for a few hundred dollars, an individual can have a custom chip made. Tiny Tapeout has now been running for four years. So what is actually being submitted? We went to the public data and counted three years’ worth.
How we did this
- Scope 19 shuttles that closed for submissions between 8 September 2023 and 7 July 2026 — 3,625 designs in total
- Source The index Tiny Tapeout publishes on GitHub (TinyTapeout/tinytapeout-index)
- What we measured Number of designs, submitters, design size, presence of analogue circuitry, process node, and genre
- On classification Size, submitters, analogue pins and process are taken straight from the index as recorded. Only genre involves judgement on our part: it is derived from words appearing in each title and description. The full criteria are listed at the end
Shuttles whose closing dates are not published (such as Cadence 25a, with 257 designs) are excluded. Where the published tables and the index disagree on design counts (TTIHP25a: 547 published, 564 in the index), we have used the index figure.
First: how Tiny Tapeout actually works
Before the numbers, it is worth being clear about the scheme itself. Get this wrong and the rest will not make sense.
Tiny Tapeout was started in 2022 by Matt Venn in the UK, as a spin-off from his Zero to ASIC Course.
What is a multi-project wafer (MPW)? Chips are fabricated in batches on silicon discs about 30 cm across, called wafers. Dedicating an entire wafer to one design costs a small fortune. An MPW puts several people’s designs onto the same wafer and splits the bill. Universities have used this approach for decades.
Tiny Tapeout takes that sharing one level further. It buys a single MPW slot, subdivides it into small plots called “tiles”, and sells those to hundreds of participants. A slot costing $10,000 split three hundred ways comes to a few tens of dollars each.
This matters: Tiny Tapeout has charged money from day one. It is not the same thing as the free programme Google funded, which appears later in this article. Tiny Tapeout is designed to cover its costs from what participants pay.
How the price has moved
For Tiny Tapeout 3 in 2023, submitting a design cost $25; receiving a physical chip and demo board cost $100.
Then the number of chips Tiny Tapeout could obtain from Efabless for the same slot price dropped from 300 to 100, because the package type — the black component the silicon sits inside — changed. Since every participant receives their own chip, the quantity needed did not change. Getting the same number now meant buying three times the slot space.
Prices were revised for Tiny Tapeout 6 in 2024: $150 for the first 100 individual submissions (one tile, chip and demo board), $300 thereafter and for companies and universities, $50 per additional tile, and analogue I/O pins from $40 each with a two-tile minimum.
Submissions did not fall off after the increase. Within the period covered here, TTSKY25b drew 316 designs, TTSKY26a 289 and TTSKY26b 273. The reason the packaging options changed has not been made public.
The front door is a browser
One more thing shapes who takes part. The entry point is Wokwi, a browser-based tool for building and simulating circuits. Originally built for Arduino work, it was reworked for logic design by Matt Venn and Uri Shaked. You can submit a design by wiring up logic gates in a browser, without writing a line of a hardware description language. The volume of practice pieces we are about to see is not unrelated to this.
And the shuttles are not all alike
One caveat before the charts. Tiny Tapeout’s shuttles do not all run on the same terms. They differ in at least three ways.
| Difference | What it means |
|---|---|
| Production vs. test runs | Shuttles ending in 0p2, 0p3 or 0p4 are test runs, used to prove out support for a new process. They draw few participants, and chips are not always distributed. These are shown in paler colours in our charts |
| Process node | Three lines run in parallel — SkyWater 130nm, IHP 130nm and GlobalFoundries 180nm — with different capabilities and different prices |
| Analogue support | Whether analogue designs are accepted varies by shuttle. And analogue work cannot fit in a single tile: it requires 1×2 or 2×2 at minimum |
Source: individual shuttle pages and analogue specifications at tinytapeout.com.
So the 19 shuttles cannot be compared like for like. We have limited ourselves to measures that do travel between them — design counts, submitters, size and analogue use — and marked the test runs by colour. As you read on, keep in mind that a good deal of the variation between shuttles comes from the terms they ran under, not from the people submitting.
1 Shuttle sizes, and the break in 2025
Lining up the 19 shuttles by closing date, two breaks stand out.
SkyWater 130nmIHP 130nmGlobalFoundries 180nm Paler bars are test runs
The first comes straight after TT09 (369 designs), which closed in November 2024. TT10, which was to follow, was cancelled the day after submissions opened. Efabless, the American company handling fabrication, shut down in March 2025.
The second is TTIHP25a (564 designs), four months later — the largest shuttle in the three years. It was a rescue run, moving stranded designs across to Germany’s IHP process.
From 2026 onwards, blue, orange and green alternate. Three processes are now running side by side.
2 Almost everyone submits exactly once
Count the submitter names and, on every shuttle, between 85 and 97 per cent of them have submitted a single design.
The proportion barely moves across three years. Shuttles tripled in size, and still the crowd kept turning over. On TTSKY25b, 316 designs came from 293 people, 97 per cent of whom submitted once. Among the handful who submitted more are people involved in running Tiny Tapeout itself; the highest count was eight.
3 Most designs are the smallest size available
Tiny Tapeout sells space on the chip in units called tiles. One by one is the smallest you can buy.
A single tile is roughly 0.16 mm by 0.1 mm. It varies a little between shuttles, but on SkyWater 130nm it is around 160 µm by 100–112 µm. A human hair is about 80 µm across, so picture two hairs by one. That space holds roughly a thousand standard cells. At the time of TT08, one tile cost $50.
The figure swings between 43 and 86 per cent. Larger shuttles tend to have a higher share of single-tile designs: 86 per cent on TT09 (369 designs), 84 per cent on TTSKY25b (316 designs). Sparsely attended test runs go the other way — TTGF0p3 sat at 9 per cent, with an average design occupying 3.59 tiles.
4 So what are people making?
We broke down TTSKY25b (316 designs), the most recent full shuttle, by genre.
“Exercises and basics” accounts for 54 per cent: template copies, and designs consisting of nothing but logic gates, counters or seven-segment displays. Many came from workshops — a technical high school in Austria, Johannes Kepler University, a University of Toronto hackathon, Hackaday Supercon.
What people are actually building
Genre buckets are abstract; the specifics give a clearer picture. We counted how many of the 316 designs mention each of the following in their title or description.
| What was built | Count | What it is |
|---|---|---|
| Logic gates | 54 | AND, OR, XOR and friends, wired together. The most common of all |
| Seven-segment displays | 25 | Driving the figure-of-eight numeric display |
| Clocks and timers | 25 | Counting seconds, showing the time |
| Counters | 19 | Circuits that count up from zero. The most basic building block there is |
| Games | 15 | Noughts and crosses, 2048, Mastermind, Pong |
| VGA video output | 14 | Driving a display directly. A popular subject |
| Binary conversion | 13 | Converting binary to decimal and similar |
| Sound and synthesisers | 11 | Recreations of classic sound chips, MIDI synths, drum machines |
| Flip-flops | 11 | Storing a single bit. A staple of coursework |
| PWM | 10 | Generating the signals that dim an LED or vary a motor’s speed |
| Crypto and hashing | 10 | Hash functions, authentication circuits, random number generation |
| Ring oscillators | 9 | An analogue-leaning subject, also used to characterise a process |
A design may appear in more than one row, so the total exceeds 316.
At the top of that list — logic gates, seven-segment displays, counters, flip-flops, adders — is the opening few chapters of any digital electronics textbook, rendered in silicon.
And yet the same 316 designs include a RISC-V SoC that boots Linux, an 8-bit SAR analogue-to-digital converter, an on-chip bandgap voltage reference, reimplementations of the classic AY-3-8913 and YM2413 sound chips, a TV-B-Gone rendered as an ASIC, and a design whose layout is itself the artwork. Textbook exercises and professional engineering, sitting on the same wafer.
CPUs and processors came to 22 designs, or 7.0 per cent; across the full three years, 9.0 per cent. Building your own CPU is the emblem of this field, but as a share of what gets submitted it remains a minority pursuit.
5 Whether you can submit analogue depends on the shuttle
Here is how many designs used analogue pins, shuttle by shuttle.
SkyWaterIHPGlobalFoundries
What this chart shows is not a rise and fall in the number of analogue designers. The line moves because it tracks whether each shuttle accepted analogue submissions at all.
- SkyWater 130nm 23 analogue designs appeared for the first time on TT06 (closing April 2024). Before that, none. Since then it has run between 4 and 21 per cent
- IHP 130nm Four designs in three years. Essentially flat at zero
- GlobalFoundries 180nm On TTGF0p3, which closed in July 2026, 16 of 32 designs — half — were analogue. It was a test run for analogue support
So the chart maps when and where the door opened, not how many people were waiting outside it. There is also a cost floor: because analogue cannot fit in a single tile, the entry price at least doubles. That is likely part of why the share stays low.
6 March 2025: what happened
Now to the gap marked “Cancelled” in Figure 1.
Underneath Tiny Tapeout was another arrangement
Tiny Tapeout charges for its service, but the fabrication behind it was handled by chipIgnite, run by the American company Efabless. Tiny Tapeout bought slots there and parcelled them out. Tiny Tapeout sat on top of Efabless.
Efabless was founded in 2014 in Palo Alto, California. Its business was standing between designers and foundries, aggregating small orders onto shared wafers.
It began with free fabrication, paid for by Google
The story starts in 2020, when Tim Ansell, then at Google, announced at an open-source conference that SkyWater Technology’s 130nm PDK would be made public.
What is a PDK? A process design kit is the set of data you need in order to design a circuit for a particular fabrication process at a particular factory: transistor characteristics, the geometric rules governing what you may draw, verification rules. Normally you cannot obtain one without signing a non-disclosure agreement, because it describes the inside of the factory. Publishing one as a free download was the news.
Alongside the PDK came a scheme to fabricate the resulting designs at no charge: the Open MPW shuttle. SkyWater provided the fab and the PDK, Efabless ran submissions and design tooling, Google paid for fabrication. The Linux Foundation’s CHIPS Alliance later joined in running it, and GlobalFoundries published a 180nm PDK and joined too. This was never one company’s project; it was several organisations dividing up the work.
Six shuttles ran by mid-2022, fabricating 240 designs out of 364 submitted. Around 60 per cent of submissions came from people who were not integrated-circuit specialists. This period produced the foundations still in use today: an open PDK, free fabrication, and Efabless’s design automation toolchain, OpenLane — which turns a written description of a circuit’s behaviour into the drawing files a factory needs.
Tiny Tapeout began in 2022, just as those foundations were laid. It was not part of the free programme, though: it started as an independent, paid service running on the same Efabless machinery.
Where the free programme ended
Here we set out only what has been made public.
In November 2023, Google announced it was transferring the SKY130 and GF180MCU PDKs to the Linux Foundation’s CHIPS Alliance. The stated reason was to move to a governance model in which industry, academia and the community could participate more broadly, opening the possibility of larger shuttle programmes with multiple sponsors. The same announcement noted that low-cost fabrication remained available through Efabless’s chipIgnite and through Tiny Tapeout, and pointed to a GF180 run, MPW-1, closing on 11 December 2023.
That announcement transferred stewardship of the PDKs. It did not declare the end of free shuttles, and it did not use the word “withdrawal”. What can be stated as fact is that no further Google-funded free shuttle has been announced since GF180 MPW-1. Free fabrication effectively stopped at that point, as far as we have been able to establish. No financial reason was ever given, and the cost of running the free shuttles has never been published.
Tim Ansell, who drove the PDK release, later left Google. He has said publicly that he pushed GlobalFoundries to run an equivalent free programme, without success. The company he went on to found, wafer.space, is now one of Tiny Tapeout’s fabrication routes — as we will see.
Efabless ran out of money
With the free programme gone, Efabless leaned on the paid chipIgnite service, priced from $9,750. In 2024 it added chipIgnite ML for edge AI work, and Chipalooza, a competition for analogue circuit design.
In January 2025, chief executive Mike Wishart used a new-year blog post to describe a bright outlook, noting that more than 50 academic institutions were using the company and that 80 commercial designs had reached fabrication since its founding.
Six weeks later, over the weekend of 1 March 2025, the company announced it was shutting down.
What was reported
Coverage clustered in the two days that followed. In order:
| Date and outlet | What was reported |
|---|---|
| 2 March eeNews Europe | Reported that the Palo Alto company had been struggling to raise funds for months, quoting Wishart: despite the company’s best efforts it had been unable to complete its latest funding round, and operations were suspended until further notice |
| 2 March Tom’s Hardware | Reported the company’s citing of “funding challenges” and that TT08 and TT09 were left in limbo, adding that whether the company would return was unknown |
| 3 March Hackster.io | Noted that Wishart had described a bright future less than a month before the closure, and reported that sources suggested a grant the company had been counting on did not materialise. This was attributed to “sources” and not presented as established fact |
| Same period SemiWiki | Carried Wishart’s message to the community, in which he described the company’s pillars as community, business models based on shared risk and reward, and open source, and thanked GlobalFoundries, SkyWater, Synopsys, Google, XFAB, the US Air Force Research Laboratory, Arm and others. It contains no account of the company’s finances |
| 13 March TechFactory (ITmedia, Japan) | An analysis piece by Yusuke Ohara, attributing the collapse to a structure in which open-source work generates little return for those who produce it. Not based on interviews with the company, but reasoning from public information |
Trade outlets elsewhere framed it as a failure to reach profitability amid rising costs and competition, but without citing financial data.
And then nothing
Seventeen months on, we could find no reporting that went to the people involved and established why. The articles above all rest on the company’s own statement. What the grant was, why the funding stopped, what the finances actually looked like — none of it has been reported.
Two explanations suggest themselves, though both are our speculation. One is that few journalists cover this field continuously: open-source silicon is too small for the semiconductor trade press and falls outside what enthusiast outlets typically cover beyond product launches. The other is that a collapsed private company leaves no door to knock on. The press office is gone, the executives do not speak, and the statement becomes the last primary source.
Efabless had run on investment since its founding in 2014, raising $13.8 million over eleven years. The most recent round was $6.3 million in October 2023, from GlobalFoundries, Synopsys and others. Around eighteen months later, unable to raise more, it stopped.
Whether the company was profitable cannot be established from outside. Efabless was privately held and under no obligation to publish accounts. Explanations for the collapse differ between outlets, but none of them, as far as we can establish, rests on the company’s financial data.
The damage
TT08’s 135 designs were awaiting packaging; TT09’s 369 were still inside SkyWater’s fab. TT10 was cancelled. Around 500 designs were left in limbo. Tiny Tapeout’s founder announced refunds.
TT08 eventually shipped in December 2025. As of writing, TT09 has no shipping date.
7 In Europe, a scheme that has run for thirty years
All of the above happened in the United States. Europe has an arrangement of quite different origin, and a much longer history.
EUROPRACTICE began in 1995 and still serves over 600 European universities and research institutes a year. It offers fabrication on shared wafers, affordable access to design tools, and training, across more than 90 process technologies. In February 2026 a consortium of five organisations led by imec announced it had secured funding from the Chips Joint Undertaking, an EU public-private partnership, through to September 2028, with a commitment to train more than 650 people annually.
Tiny Tapeout’s IHP shuttles sit in this same context. Porting the flow to IHP’s PDK was funded by the Swiss SwissChips initiative, and the wafers themselves were paid for by a German Federal Ministry of Education and Research project. For TTIHP26a, closing in March 2026, SwissChips sponsored the entire shuttle, making it free to submit for anyone resident in Switzerland.
| What happened in the US | How Europe does it | |
|---|---|---|
| Who carries it | A venture-backed company (Efabless), plus funding from large firms | Public money (EU Chips JU, Germany’s BMBF, Switzerland’s SwissChips) and research institutions |
| Duration | The free era ran roughly three years, 2020–2023 | EUROPRACTICE has run since 1995 — thirty years |
| When it stopped | Operations ceased in March 2025; some 500 designs stranded | Funding secured through 2028, announced February 2026 |
| Who can use it | Anyone, provided the design is open source | Principally European universities and research institutes, with some free places for residents |
The two were built for different purposes and different users; this is not a ranking.
What to make of that contrast is a matter for the reader. What can be stated as fact is that the shared-shuttle model has always depended on money arriving from somewhere other than participants. In the United States that money came from venture capital; in Europe it comes from the public purse. The first stopped after eleven years. The second has run for thirty, and is funded to 2028.
8 The gap closed within eighteen months
Two of Efabless’s founders set up ChipFoundry.io, restoring the route to SkyWater 130nm. The price rose from $10,000 to $15,000 for 100 parts, with a new condition: the operator reserves the right to postpone a shuttle that is less than half full.
Around the same time, wafer.space launched and Cadence began running its own SKY130 shuttle. OpenLane became LibreLane under FOSSi Foundation stewardship. And because Tiny Tapeout had already been working with IHP through the SwissChips initiative, it was able to move most of the stranded designs onto TTIHP25a.
As of August 2026, Tiny Tapeout runs three fabrication routes in parallel: SkyWater 130nm through ChipFoundry, IHP 130nm, and GlobalFoundries 180nm through wafer.space. The dependence on a single supplier was resolved within eighteen months of the collapse.
Appendix: limitations and classification criteria
Genres were assigned by lower-casing each title and description and searching for the following terms in order; a design falls into the first category it matches, and every design falls into exactly one.
- Exercises and basics Terms indicating a template or coursework origin: template, test, wokwi, workshop, my first, hackathon. Also counter, flip-flop, adder, 7-seg, logic gate, led, binary and similar, where the design is a single basic circuit
- Analogue and signal processing A non-empty analog_pins field in the index; or analog, adc, dac, opamp, oscillator, pll, bandgap
- CPUs and processors cpu, processor, risc-v, soc, turing, gpu, instruction set
- Comms and interfaces uart, spi, i2c, axi, serial, morse, hamming, encoder, decoder
- Video, audio, games and art vga, video, game, synth, audio, music, display, clock, generative
- Crypto and arithmetic crypt, hash, lfsr, multiplier, fft, filter, pwm, alu
- Memory and logic blocks rom, ram, fsm, state machine, shift register
- Other Anything matching none of the above
Three limitations worth stating plainly.
- We matched only titles and descriptions. We did not read the source code or examine the circuits. Designs were sorted mechanically according to whether the name and short description given by their author contained our search terms. Two functionally identical circuits can land in different categories depending on what their authors called them
- “Other” does not go away. Across the full three years it accounts for 20.0 per cent. Earlier shuttles carry more idiosyncratic names and match fewer terms. On the recent TTSKY25b it was 5.7 per cent
- The shuttles ran on different terms. As set out at the start, test runs and production runs differ, as do processes and analogue support. Some of the movement in these figures reflects the rules, not the people