Frederic Jacobs🇳🇱🚲🌷& 🍪 A trip to the Netherlands is never a wasted opportunity to nerd out on semiconductor technology. A thesis defence brought me to TU Delft, but this time I was not gonna miss the opportunity to see the Mapper machine!
🎬 Story time!
Back when shooting pulsed lasers on a thin droplet plasma requiring mirrors that would be no thicker than hair if they were spread on the size of Germany, and laser accuracy allowing you to shoot a golf ball on the moon (aka EUV lithography) was considered to be mad science, alternative lithography improvements were being developed to continue the relentless demand for smaller and more efficient transistors.
Mapper, a TU Delft Spin-Off, was promising cheaper, higher throughput, and maskless lithography.
The machine is very elegant and demonstrated a solution that didn't require masks, so you could take your circuit file and just print the circuit with one machine, allowing to do even very small batches of chips. The machine is really amazing to see and is in the Applied Physics building of the university.
The machine is very elegant and demonstrated a solution that didn't require masks, so you could take your circuit file and just print the circuit with one machine, allowing to do even very small batches of chips. The machine is really amazing to see and is in the Applied Physics building of the university.
The company secured some key orders such as Global Foundries that works to tape out critical US DoD chips that end up in things like fighter jets or missiles.
But at the end of 2018, the company was running out of money. Making lithography equipment is very capital intensive and the machines were not selling fast enough. Mapper was looking for a buyer.
Prospective 🇨🇳 buyers were showing up, and that freaked out Washington.
The Pentagon contacted the 🇳🇱 Ministries of the Economy and Defence, the message was clear: "China can't buy Mapper”. This was escalated to Mark Rutte over the Christmas holidays of 2018.
The Dutch government, under US pressure, required ASML to buy Mapper. Since ASML had no interest in multibeam & was all-in on EUV, the Mapper engineers were re-assigned to metrology projects, inspecting the chips instead of producing them.
ASML still operates a small office out of Delft, which happens to be at the ex-Mapper office.
There were only 2 working machines produced, one of which is the one that was rescued by Kampherbeek and TU Delft professor Kruit, and can be seen on campus.
Today, it occupies an important role in Dutch history as the moment the government woke up to the critical role its companies were playing in the geopolitical world of chip production 🍪
The years since have only exacerbated this further
🔚
I thought that was the end of the thread but @jmovs clearly convinced me that everything I read so far on the Mapper was underselling its achievements 🤯
https://mastodon.social/@jmovs/114699887334170332
https://mastodon.social/@jmovs/114699887334170332
bert hubert 🇺🇦🇪🇺🇺🇦@fj fun fact, the Mapper founders are friends of mine and if things had worked out a little bit differently I might have well ended up as employee 5 or so.
Mike OlsonThe fact that the Netherlands, via ASML, utterly dominates high-end lithography for chip production is amazing and pleases me.
It also hammers home the importance of close international trade alliances and the foolhardiness of the "reshoring" efforts for industry generally and chip production in particular. Much longer rant than I have space for here.
Was unaware of Mapper. Next time I am in Delft I may see if I can visit! Thanks for the background and pics.
Justin Scholz@bert_hubert @fj oh cool! BTW thanks @bert_hubert for the initial retweet - it made me go on a tangent (in case you missed my reply thread) :)
Leon Overweel@fj thanks for sharing!
Sobex
Ölbaum
Thijs Alkemade@fj My dad used to work in that building. 🙂 He wasn’t involved in this project, but he did work on nanoscience. Getting to operate an electron microscope while visiting him there was really cool.
@xnyhps Lors of cool stuff happening on engineering campuses 🪄🧙
@fj not sure what you mean with „didn‘t require wafers“. It didn’t require masks but we always used wafers (I worked 2018 at Mapper :) )
@jmovs Typo! Good catch, updated!
Must have been a fascinating time to work at the company 🙃
Must have been a fascinating time to work at the company 🙃
@fj oh yeah it was.
I‘m not sure I would qualify the mapper solution as „elegant“ - I would say it was its own kind of batshit crazy. And the mapper engineers were some of the smartest people I ever met. 250 people achieving something equivalent of what ASML took 10.000.
@fj from its origins to then arriving at a system with a custom designed vacuum chamber, custom designed wafer stage, custom developed and self produced MEMS chip in a Russian fab (why that was built in Russia is beyond me), custom developer FPGAs to stream the pattern for printing on the chip because back then computing chips were not fast enough. Custom developed electron source with custom developed electron optics all running on of course super intricate custom software.
@fj and don‘t get me started on the craziest configuration and document management system I ever witnessed 🤣
@jmovs You mean Word documents being emailed don’t you 🥲
@fj NO!
THERE WERE 0 documents emailed inside mapper. I‘m not kidding.
Instead there was „CMT“ - configuration management tool. A simple website with a simple upload button and a search bar. Every file had a unique URL, and you could upload newer revisions of a file and link to either the lasted revision or a specific one.
Literally anything was always in CMT. NOTHING was ever emailed.
@fj You might ask: „How were the files organised? Did you use folders or sth?“
Good question. No folders. None.
Instead, there were design files and a strict naming scheme. And for the machines, the design files told you how things to be done and there were build lists that then gave you info on a specific info. Everything was then documented in Excel files that then linked to the other files (all through CMT unique links).
@fj the most jarring thing in the beginning definitely was „no folders, no start page, no emailing files, no. You start with a build listing which you can see according to which design documents which subsystem was built with which build list“.
It was really cool once you got the hang of it.
@fj if you want to know more about- I‘m happy to share more about this one haha.
@fj aside from this, Mapper had one system operating at LETI France - not sure what happened with that machine. And another one in Delft I think. Globalfoundries I don‘t remember but I think TSMC also tested a machine but way earlier.
@fj ha, and a MEMS element that was planned to contain 15000 holes, where each individual hole had two sides of a capacitor around, one connected to a light diode, to which a glass was connected. This would then bundle into the „fiber tree“ that went to the FPGA rack.
You had to use light because you can‘t use electricity when you‘re soothing electrons at this MEMS element. Once the beam went through these holes, you could control every single hole individually whether it would deflect or not.
@fj and the beams would continue and then either go through a separate hole sheet and hit the wafer that was moving or be going into the beamstop.
There is a great video about that part on YouTube: https://youtu.be/H8RA5QYT9jQ
System Architecture of the Pattern Streamer for Mapper Lithography
@fj here, it‘s not the mirror being so thin but the surface roughness being so smooth I think. :)
@fj another typo: „golf ball“ :)