*Why bother with the rotation stage?* Good question! It was for multi-layer alignments. I.e. pattern first layer wherever on the substrate, and then align all subsequent layers to registration marks on the first layer. In theory this could all be done in software (rotate the image digitally) but it introduces a lot of complexity to the software. I thought mechanical solution would be easier. But my design (direct drive to a stepper motor) was a flawed idea and in retrospect there are much better ways to do it.
The easy way to make a rotary vacuum too head is to use a dual hollow shaft stepper motor with a flexible silicone pipe at the bottom end and the tool head at the top. Obviously you have limited travel but you can easily do 360 degrees. That is how DIY pick and place machines work. No need for o-ring seals.
Honestly, when I first saw you assembling the rotation stage my first thought was "how will he ensure the rotational stage is planar?" Based on your DOF with that level of magnification, I assumed that traditional indicators you'd see in a machine shop wouldn't be good enough to adjust it into focus across the whole stage. It's an unfortunate problem to run into, and besides having better methods of measuring to improve the accuracy you would either need some different equipment (surface/rotary grinders/etc) or a lot of patience and skill with sanding/lapping.
or you could locate the rotation by removing a part of the circular wafer, like it is done in industry. If you have a straight edge you should just need a couple of dowel pins to locate against. (in theory)
@@BreakingTaps to do layer-to-layer, you'll need to be able to figure out your pixel size in X and Y, as well as the skew of your optics. Hint: calculate your per pixel H move with X and Y, then do the same for V pixels. You'll have 4 calibrations numbers that you should be able to do all of the math with after. Also, you'll need to deal with your flat field correction for the objective. You can either shrink the exposure window or you can grayscale your output and correct with a digital mask. The easiest way is to use a white background and capture a blank window with the red wavelength with your inspection camera. You'll then need to map camera pixels to exposure pixels
it still took all those minds, and many more for the intermediate history that has made the task this attainable. It's really inspiring in so many perspectives
We are standing on the shoulders of men, who are standing on the shoulders of men... dating back to the first dude who rubbed two sticks together and thought, "huh... Neat"
Greeting's from Intel's sub-nm research team! It's super incredible to see hobbyists getting to the point of making real semiconductors in the home shops. I did my PhD on reducing resistance in chip-to-chip bonds, so if you ever get to that point where you want to make a really tiny sandwich, definitely get in touch down here!
Is that sub-nm (A7 process node)? What are some recent public things representative of the kind of work your team does? The problems routinely solved in the semiconductor industry are incredible. This hobbyist level stuff is really impressive, but EUV lithography systems feel less and less real the more I find out about them. I'd love to have a few billion of your sandwiches :P
@@reddragonflyxx657 Sub-nm referes right now to any node from Intel ending in A, as those need Angstrom measurements. 20A/2nm is the first, followed by 18A, then likely 14A. Most of the work from my team has been in the theoretical side of things, but you can see the first applications of our work in the Intel4 and Intel3 nodes. The 20A die of Arrow Lake will be the first with PowerVia, which will also mean it is the first designed with tools my team developed for sub-nm.
@@atomicrevenger4745 What are you on about? Silicon MOSFETS (for digital logic) with critical dimensions of ~3um are over 4 decades old. If you meant 3nm, then it really won't matter anytime soon, since the upcoming "sub nanometre" nodes are still expected to have a minimum inter-transistor gate pitch of at least 30~50nm, that's *10~15x* bigger. Also, while it's true that shrinking planar transistors (and even finfets) down below a gate pitch of around 20nm brings exponentially increasing static leakage and decreased gate control (think short-channel effects, like DIBL), silicon doesn't just magically "lose its semiconductor switching properties" at 3nm and in any case, we are unlikely to see any Fab actually get anywhere near that level of density during this decade. The industry-wide switch to GAAFETs might, eventually, even allow for the introduction of commercially viable process nodes, with actual inter-transistor gate pitches in the 5~10nm range; however it's still far to early to say much with certainty about how GAAFETS will mature.
"Screws take up physical space" is such a funny cad-to-IRL problem. I made a similar mistake with a lego robot arm I was designing in cad, and everything was very meticulously physically simulated to ensure all joints, cables, etc were perfectly fine. And then I tried to build it and realized a few of my critical fastening pins along the arm's base were unable to be inserted because they would have to simply cease to exist and then magically re-manifest inside the already-assembled part, whoops.
McMaster-Carr supplies 3D models for most of their fasteners, making it easy to import in fasteners so you don't need to model them yourself. If you have a lot it can definitely start eating up memory, so you can leave them hidden/suppressed for a lot of the design process, but having them in the model is really useful. McMaster also has add-ins for Solidworks and Fusion360 to directly search/import models instead of having to download from a browser then import.
At least you learnt from your own experience here. As a welder fabricator I have PTSD about fully engineered designs that were physically impossible to weld lol. It would be upto to us to make modifications so it would work, and fully document those changes so the engineer could approve or revise them.
Anybody can make anything with enough interest, time, research, and resources, but an EE can design something that your insurance company will let your employees operate. "I'm not an EE" is basically a line of distinction like someone saying they aren't giving medical, financial, or legal advice.
i just realised something, in aerospace theres Aerobraking, and the cursed euphemism Lithobraking, which just means crashing into the ground, but if you use the terms the same way, Lithography literally means "Rock Writing" we write patterns into rocks to make them think... literal runic carvings, we are magical rune carvers
If my memory is correct, it's called "lithography" because way back in the day (1800's), they used acid to etch patterns into limestone, and that patterned limestone could be used for as a printing block (basically, a big stamp) to repeatably print custom patterns onto paper.
Hey! Optical scientist here! I design and build these kind of setups for living (used to be custom microscope systems, but now space instruments). So fun to see what you can achieve with DIY stuff! Youre doing an excellent job!! For me building and aligning is always the best part of a project. I think you can ease your alignment process a lot by using good alignment tools and mounts with the appropiate degree of freedom control. For instance, putting the objective on a z-translation cage mount would massively improve the alignment control. For the alignment tools I would suggest cheap ones screw on alignment targets and frosted glass disks. You can also make these yourself! When I design and build systems I spend most of my time thinking how I will align it. Where is fine adjustment needed? What can be mechanically toleranced? Thinking ahead of all these things saves you massives amounts of time. I know for sure your next optical system will be even better! Keep it up😊
@@MakeKasprzak that's only if the parts work: what usually happens is that you waste time and money, and then you have to buy the proper parts anyway. Or even worse, you go out of budget and can't complete the project at all...
@@Beregorn88 That's not the fault of the 3D print though - that's the fault of the optomechanical designer. Using the wrong process for the job does not discredit the process itself. I absolutely think 3D printers have their merit in the prototyping process, also in the optical lab - you just need to know when it's the right tool for the job.
Right?! Huygens Optics is like the equivalent of that lesser known restaurant all the best chefs love to go to after work. I see all my favorite youtubers in the comments over there! Super channel!
So I work as an engineer developing lithography machines and it is funny to me how I have very little knowledge of the process as a whole, while noticing a lot of small things that would never fly at work. It is great to see someone build a working machine in an hour instead of focusing on a small gear in a very large machine for a year at a time.
Working on v2 so I'm all ears if you have tips! Top priorities are: a precision base for the stage (granite, optical breadboard, etc), reducing stack height of the stage to reduce abbe errors (and fix that stupid rotation stage), and re-arranging the optics so they aren't hanging/cantilevered or attached to the frame
Or got killed by ESD, because those bare UV dies are really sensitive to it as well. Best to also add in the protective diodes that the manufacturer recommends, and which are almost never used. Shorted diodes either ran too hot, because of poor thermals, or got zapped by ESD.
@@SeanBZA There was a heatsink right next to the diodes - are those diode driven or from other nearby equipment? Might want to install temp sensors and/or increase heatsink capacity and/or airflow.
I think the main problem w.r.t the LEDs is at the 8:50 mark where the LEDs are being soldered. Note how they "soldered up a bit wonky", which might have seemed like a cosmetic issue for the most part at that moment. However, it has severe implications for the thermal performance of the system. Basically, the LEDs can't properly dissipate the heat they generate. As a result, their lifetime is shortened considerably. Running them at lower current will help some, but they'll still die prematurely. The real solution is (1) ensure proper dosing of the solder paste. There only needs to be just enough to flow the entire contact surface, but it shouldn't bead up, raising the LED away from the copper PCB. The other part (2) is to ensure that heat dissipation from the copper PCB is as efficient as possible. I'm going to assume that (2) is dealt with because apparently this sub-assembly was lifted from an existing/functioning machine. However, (1) still needs to be addressed. I've dealt with this exact same problem with high-power SMD LEDs many times, so the above is my best guess based on 1st-hand experience. It doesn't discount other possible causes, but I'd really start tackling the problem here. It's a simple fix, in principle. That's the good news.
@@leonordin3052 More details on the Zaber page, but it's a crossed roller bearing stage. The absolute positioning accuracy and repeatability is in large part from the encoders, but the pitch/yaw/roll accuracy are all from mechanical design of the stage (i.e. encoders can't correct for a pitch error in the axis, that's purely mechanical/assembly) (although since it comes with interferometry data you can compensate for pitch/yaw/roll errors in software, will touch on that in the next video)
3 місяці тому+46
The things you cut from this are my favorite parts - the really nerdy stuff. I care a lot less about watching the assembly of the frame, than I do about the design of the frame for instance. I enjoyed this video, but I'm _really_ looking forwards to the next, regardless of the effort that goes into it.
Yeah sorry about that! The original plan was half assembly, half design discussion. But the decision to build a second machine wrecked my video-timeline-planning and it was hard to do both. At least this way I can spend gobs of time talking about stuff like abbe error compensation in the next video :)
Fun fact about standing waves in photolithography, they are worse the more narrow your source wavelength filters are. Standing waves are worse for laser and led sources vs hg lamps. Source: I make antireflective coatings for litho for a living.
Ah yeah, that makes sense! I assume a wider bandwidth will "smooth" the standing wave since the different frequencies all interfere at slightly different locations? I was impressed how crisp and "step like" the standing waves were! I had seen them in the literature before but never in my own tests. A quick post-exposure bake seemed to get rid of the worst standing waves luckily :)
@@BreakingTaps You have your work cut out for you! Early in my career I was also responsible for alignment and overlay matching a fleet of ASML tools. So many variables!
1 micrometer lithography. I am impressed. Really makes me wonder and appreciate how much effort and alignment goes into nanometer printing with lithography !!!
I've been working with a friend on a high resolution film scanner design for the past year, and I can tell you I powerfully empathize with all of those "I have to totally disassemble this and then reassemble and realign it moments and I would rather walk slowly into the ocean" moments. It's a real rollercoaster. BTW I'd spend the time to either figure out how those LEDs failed, or make them easier to replace. LEDs that failed once in a given installation are pretty much guaranteed to fail again.
Enjoying this greatly. I chuckled at the comment about professional use of 80/20. You might be surprised to learn how much 80/20 or Rexroth is used in production multimillion dollar tools by a variety of tool vendors...
Finally got a chance to finish watching this. I wanted to add that again, even on multimillion dollar production 300mm tools with the very nice precision motion components......there's still hundreds of hours spent during the life of the machine fiddling with 0.05mm shims.
Have you considered using a blu-ray pickup for both the UV laser and optics systems, and as a stage? A standard pickup has a nanometer accurate voice coil holding a lens, and a UV laser, as well as a photodiode, through you probably don't need that. I know there's been a paper on using them as a micro 3d resin printer. The idea is you use two pickups facing each other at right angles, since they only move in one horizontal axis as well as vertically, so you attach the sample to one pickup and use the other to scan the laser over the sample. I've used them to make a laser scanning microscope with extremely high magnification (diffraction-limited), so the potential is there. You can even use the photodiode to measure the distance to the sample with nanometer accuracy, which should help keep things in focus. The real limiting factor is that the pickup's coils, while extremely accurate, can't hold much weight. That means if you use one to hold the sample, you'd be limited to a couple grams max. Fun fact: you can even use those same pickups to make an atomic force microscope
@@mr_gerber I tried posting the link from my phone, didn't go through so I'm trying on desktop (update, still didn't go through...). Here's the link for the project (just the project location relative to giithub's home page in case the issue was from posting a link): NilanEkanayake/LaserScanningMicroscope . It's currently not very polished, and will likely have bugs and require a decent amount of knowledge on the subject to build. It also uses a DVD laser (~650nm).
I mapped my z height using interferometry and DFTfringe and adjust my heights with a piezo linear actuator. X and Y were mapped using a reference mirror tilted at 45deg and about 12 feet to calculate any wobble (it showed that i had a slight twist and bow that i compensated for in software). I also dont use a lens on my camera since the light coming out of the objective is suppose to be parallel. A cheap photoresist that i use is D&S Posicoat used for metal etching. I can easily get ~800nm features with it (its as far as i could test before my daughter was born). I took alot of inspiration from hyugens optics with a mix from applied science. I have mostly made mesfets due to their simplicity. Also materials project is a great resource when trying to determine what materials to use, bandgaps, mobilities, lattices, energy levels, and compatible substrates.
Seconded, while troubleshooting uniformity I've unplugged the piezo signal on a maskless aligner during a write. Work it's able to do to keep focus cannot be understated.
Can I pick your brain about the camera comment? I have a secondhand stereo microscope from Vision and am trying to mount a camera on one of the two optics paths. I believe the light is parallel in these stacks but a camera in the path doesn't work right... How do I reason about camera sensors with parallel light beams? I thought we had to project a focused image onto the sensor?
Hrmm. I would have expected the construction of a 1um machine to look like, you know, a high end machine tool. Lapped surfaces, air bearings, granite, that kind of stuff. Machined aluminum on a warped piece of sheet metal with a bunch of shims, holy wow, getting what you did is super impressive!
It was kinda funny watching you build a 2M$ machine, yet struggling to make a simple metal box :D But I also saw that you improved massively just after the first try, so it shows how smart you are and how quickly you learn. Amazing job with the project, I hope we'll see more of it in the future!
This is something you can watch twice and still learn something. My heart goes out to you. One component can really drive you nuts if it is not easy to access, especially when you have to re align the optics. I am looking extremely forward to your next video 👍
And is cool to use! We got that very uMLA last year and oh boy it sped up my lithography experiments from one layer per day to 2-3 x3 Ability to expose in Gray-scale is fun too
Can we please all take a moment to appreciate the section (15:10 after assembly) where you explain everything you would've designed differently. There is so much to learn from there!
This is possibly my favorite channel. You have the perfect personality for doing all the things you do and talking about them, no idea how you are able to stay so optimistic and calm when I know how darn frustrating and tedious many of the issues that can and do come up are. This particular video came out at a really great time for me as I'm about to be working on literally the exact same thing (something I've been working towards for years but finally now have the time)
Loved the edit at about 26:30. I absolutely love everything you've been doing. Its wild to see just how possible it is to recreate this kind of technology in a home lab.
This is really amazing. Keep up the awesome work! I really appreciate that you show both the positive outcomes and what went wrong. The failures, setbacks, and frustrations are what I relate to the most. Most creators only showcase their progress, but seeing someone question their life choices, just like I do when working on a project (though not as complex as yours), really motivates me to keep going. So, thank you!
For the UV LEDs, you might want to check the paste you used, if it is WS type and you didn't ultrasonic it or flush it with water and often IPA after that, there's a good chance the acidic PH ate away at everything beneath them and turned it into a conductive mess. WS=water soluble, NC=no clean. WS is good because you CAN clean it up readily and remove flux residues, NC is okay for digital stuff etc, where a little bit of flux isn't too big a deal even if it can cause megaohm parasitic impedances. BUT I'm not suggesting you'd want to use NC, as the hot environment might also be enough to activate it enough to slowly attack the LEDs while they are on if they get hot enough, ideally you'd use WS and clean it off so no flux is left, it's just that WS will simply corrode it with time, and not just while it's hot. Else heat indeed fused them shorted, suggest monitoring their temperature as part of the process or upgrade heatsink and cooling.
Oooh, I didn't realize that was a thing. Just checked and it's "kester ep256 no clean", also very expired (got it a few years ago for a project that I never completed). Good to know! I'll pick up some WS for the next batch of LEDs.
7:40 there are at least 3 universities in the US with the abbreviation CMU, it is probably wise to specify which one you mean. in this case, its Carnegie Melon University, for those curious.
Congrats on the mammoth achievement! I yearn for the day when custom IC design becomes as easily accessible as PCB printing - a true superpower in everybody's hand. Surely, a step in that direction. Amazing work. Keep it up.
Cool stuff, I attempted something similar but the deposition of metals was too expensive for me so I stopped at the lithography part. Some advice, Mount everything to a granite surface plate, you can use a diamond brazed bit and a router to bore holes into the surface plate then glue inserts to secure any apparatus. This makes aligning things much easier. Not all DMD offered by TI are the same, the one you used has really poor optical performance when it come to
What do you mean you are not an engineer. You are a very talented engineer. Ability to pull this off and passion is very rare. Consider me ur new fan and a subscriber. I work in this industry and this video was fun to watch
I want a video of you partnering with Applied Science. You both absolutely stun me with your expertise and video qualities for such complex topics. I learn so much every time. Amazing work!
Just a tip: The "nose" on a stepper motor is there to align the axis where you planned. This takes away any tolerance in the holes/bolts. If you had matched the hole for alignment you could have enlarged all the bolt-holes and still had motor centered on fixture as planned. Not a big deal in this design but would have been important in a design where you rely on shaft position for gear mesh or belts etc.
this really is awesome, i have worked on Nikon Photolithography tools and now electron beam lithography tools, this really hits the basics very well. This guy is next level smart, time for him to move into semiconductor land.
Makes one appreciate the 3nm process all the more. Well done, the best bits are the mistakes for people to learn from, thanks for the candid approach to the videos.
Perhaps this is why people prefer working in the digital space. Physics is so much trouble. But I’m really happy to see your knowledge and determination on display. Stretching is edifying.
Notes on the diode test setting of the multimeter for LEDs: Standard silicon diodes will read around 0.65V, red LEDs will be closer to 2V, and blue will be closer to 3.4V. Larger voltages for shorter wavelengths, in general. There's a lot of variation between different LEDs, so treat these as rough numbers. When you measured the red LED board one direction and saw 0.66V, that was probably a reverse protection diode. I couldn't find the LED PCBs from TI's docs to check, but the Hacker-Fab board has two LEDs in series on theirs. That means that the forward direction for the red LED board should read around 4V. Depending on the multimeter, 4V is too high to be read by the diode test function, so it'll show up as overrange. The 0.071V you measured for the UV LEDs is pretty close to a short. They are definitely toast (but shorted, not open circuit). If you wanted to measure the 4V of two red LEDs in series with your meter, you'd maybe want to look for one that has a zener diode test function as that'll read up to a much higher voltage. (And I'm really looking forward to the 2nd video with the engineering deep dive.)
This is way,way beyond my reasoning and ability to logically form a conclusion to this technology.I applaud these brilliant entrepreneurs! Hopefully their efforts are rewarded and not stolen.❤
It's amazing to see that you are as bad as I am at designing mechanical parts. It gives me hope that the next step, doing something amazing, is right around the corner.
Holy crap I saw this video recommended and was like "This is legit gonna be such an awesome video" and then realized it's from my favorite creator. So cool!
You're accomplishing things I could only dream of. Doing obscure but extremely interesting things the hard way used to be my hobby. I did some truly interesting things but... with age came an essential, though benign, tremor. Looking forward to the next video. I immensely enjoy watching them.
After taking a materials engineering course, I understand just enough that I can follow your videos, but you still constantly teach me and expand my knowledge, keep up the awesome videos man!
I really can't thank you enough for showing all the annoying little mistakes, like e.g. forgetting that screws actually take up physical space. When I make similar mistakes I often start seriously doubting my higher brain functions. Seeing someone admirable like you, fall into those same pits, really starts to convince me, these mistakes are essentially unavoidable. I wish you the best of luck in your further endeavors from Germany!
Recreating lost technology is immensely difficult! It the little things (seldom documented) that are the most challenging to rediscover, things that no one at the time considered important. Keep at it you will figure it out!
Must admit, the fact that the whole "calibrating texas electronics lenses" section was filmed out of focus was way funnier than it should be. I understand having to put all the focusing mana into the primary objective though, great video
It's soooo good. 🤩 Legitimately one of the nicest pieces of equipment I've worked with. Excited to improve the machine in version 2 so that we can really push the limits of the stage.
@BreakingTaps maby try using hydrofluroic 1-2 percent to clean before addition of the photoresist for better adhesion. The age of the photoresist and how it is stored will impact the quality. I would suggest a photoresist that is newer
Shout out to zaber for just donating it. I just took a look at the pricing for those things and that is definitely not the kind of thing you just casually pick up for a project lol. Very kind of them to just donate it without the usual quid pro quo promotional stuff.
A tip for metal work from a welder (if you do more metal work): An angle grinder with a flap disc. Flap discs are like what sandpaper is for wood, but for metal. They are great for rounding/cleaning edges and getting mill scale off. It makes welding a breeze and would have taken less than 5 minutes to clean all the edges of your pieces. When I worked at a mom and pop fab shop doing odd jobs like store racks, we would use a lot of in-house cut sheet metal and we had boxes of the flap discs for edge cleaning simply because while they are quite consumable, they work VERY well.
Such vast latitude in DIY and hobbyist opportunities & capabilities is what makes a super power, not its nuclear arms stocks. Impressive job. Top notch patience and dedication. Kudos! from Kenya.
A better option vs your steel sheet metal plates would be to use waterjet and machined MIC6 aluminum plates, at least for anything that can be subsequently aligned with gauge blocks.
So awesome that you built this by the way :) I remember learning all those same lessons as a junior mechanical engineer. Also hilarious how you ended up back at building masks 😂😂😂
would be really cool to see a like 80s era transistor, opamp, whatever like all the way, as a series. so getting the wafer exposure done enough, maybe place transistors for some existing design, figure out bond wires, make packages (did they stamp the cans out?), and putting it in a basic circuit to do something
Decades back, my grandfather had a small wood shop setup in a shed. Now someone who seems to be roughly the same age as me, essentially has a cpu shop. This is equal parts awesome and amusing.
generally for such application all the components should be precisely grounded,using indicator during installation checking parallelism and straightness .xy stage should be bolted on granite plate, the whole setting a vibration isolation stage
Thank you very much for this video! I happen to work on the first in field GCA steppers. Got to help design the FAB enclosures, Fab placement and involved with the installs up until we put them into full operation mode at which point I moved on to other projects. Worked with field service techs and engineers while we tried to get them on line. Good time!!
Thanks! I want to make some image sensors (either CCDs, or the very early one-transistor CMOS sensors)! Just really neat devices and it'd be fun to record some video on a DIY image sensor. After that I kinda want to play with analog stuff, maybe some neuromorphic chips that rely on analog values etc. Basically looking at interesting devices that don't mind having kinda big feature sizes :)
It’s cool to see you using a zaber stage. I had a couple internships there during college and did testing on their first generation linear motor devices. Crazy how accurate and fast they can be.
You should *really* have read the TI documentation regarding DMDs used in the UV range, as well as that regarding your chosen chip. To cut a long story very short, you will definitely ruin (or at least degrade) that DMD if you blast it with < 400nm light. Furthermore, the optical properties at UV wavelengths of components meant for use in the visible spectrum can cause *lots* of issues, such as: - A lot of light being reflected back into the LEDs, - UV absorbtion causing extreme heating of stacked components without good cooling. The glass covering the DMD and any of the optical lenses where thermal expansion would cause a large deviation in the focus or worsening of optical abberations at UV wavelengths. Even the metallization of the DMD mirrors in those consumer-grade chips can sometimes absorb a lot of UV Iight. - Ineffective anti-reflection coatings. Also make sure to check the dispersion of your optics, or even better, use the lowest power light source with a wavelength that is as close as you can get to that of your exposure LEDs (without noticably exposing the resist), to minimize focus errors caused by dispersion. I can't think of a single good use for red light on this kind of system, so maybe use that channel for a 450~460nm low-powered illumination driver. You also need to ensure that the blue (>400nm) light that hits the DMD is properly diffused and collimated (I'm unsure how the optical train of that dev kit is configured, but you definitely can't just shine those LEDs directly onto the DMD.) TI has system design examples for each DMD that they sell to help you with this. Uneven input lighting doesn't just translate to a different dose-rate at the wafer, it also causes uneven heating and expansion of the DMD's mirrors. You could even be locally exceeding the specified absolute maximum spectral illumination intensity, thereby constantly degrading the reflectivity and/or the dynamic performance of the mirrors or the underlying CMOS latch at those hotspots, which would eventually lead to either 'dark' spots or 'stuck' pixels. I didn't see how you were driving the LEDs, but you can't just re-use the integrated drivers and need to use a good quality, low noise *linear* current regulator with the usual UV, OV, OL and transient protection features to protect the LEDs and to keep them at a constant brightness. You should ideally take great efforts to keep the LEDs at as perfectly constant of a temperature as you can manage (also keep them under 50°C, most ideally around 30°, and only drive them at ~half the recommended current/power if you want them to last for a decent amount of time.), mainly to maximize the amount of time that you can use a constant exposure time. It's no fun at all to constantly have to re-run your exposure optimization process, just because your exposure intensity keeps randomly decreasing as the components age. Or, you know, invest in a closed-loop exposure dose control system to automatically compensate for the average brightness variations at the wafer (which will only work if each LED ages at nearly the same rate over the entire projected field, but I digress). Finally, I know almost nothing about the mechanical side of engineering, but I do know that NEMA motor has no right to be directly coupled to the image plane. You should also completely discard that magnetic attachment system. Those magnets aren't going to snap back within 10um of their last position. It would be *much* better to lap the chuck against its mating surface, or by some other means generate a large, flat and even contact surface for it to repeatably and mechanically mount against.
Cheers! Some of these are definitely on my radar, just didn't have time to fully implement. Taking notes about some of the other points you brought up and will try to incorporate them in future iterations 👍
@@BreakingTaps Thanks for taking the time to read all that *and* for writing a reply! It is much appreciated. (PS. Please excuse my bad english. I do try my best, but languages have never been something that I could really get the hang of.) I would really love to see this project succeed so, if you ever think you may have any questions that an EE might be able to help with, don't hesitate to contact me. I even have some excess free time at the moment so I wouldn't mind doing some quick circuit design/simulation and/or PCB design and layout work for this project, especially if it's regarding something that I'm experienced with (obviously, I wouldn't have any IP rights to the designs and don't want to be publicly credited or thanked at all, nor would you be obliged to use any of my contributions. I just want to help you to avoid some of the common pitfalls and "traps for new players", that I'm familiar with.)
you have an amazing editing style. often times these techy/sciencey videos can get pretty boring to look at, but your 46min video was over in a heartbeat. earned a new subscriber! keep up the great work 🫶🏼
That's one of the most "real" and dignified sponsor promotions I've seen - congrats - you're doing so much right with your content. Design for maintainability next ;)
Why even have the rotation stage on the maskless system? Seems like a ton of complexity (and stacking errors...) for zero benefit when the projection can be rotated in software.
In theory it could be handled purely in software, but makes the software notably more tricky. With mostly mechanical alignment the optical/software alignment is simple. "find feature, fix minor rotation error, display pre-cropped image". But if you want to allow arbitrary rotation you need to identify where you are, translate from mechanical space to image space, rotate and crop out the relevant portion from a huge pattern (my test images were like 50,000 x 50,000 pixels before getting cut up for stitching), display and hope all your calibration is dialed. Totally doable, just a different set of challenges. I gambled that the mechanical side would be easier/faster :) That said, v2 is definitely going to have a different design. I made some silly choices here :)
@@BreakingTaps Yeah, I guess I'm approaching this from the point of view of mainly a software engineer :) I will likely be working on maskless photography soon-ish, and my thoughts are that since the 70s/80s two technologies have really moved forward: the existence of DMD/DLP (which you covered) and availability of compute power/algorithms. Meanwhile, machining precision elements is still expensive. And you see reliance on software solutions in manufacturing elsewhere, too - imagine modern PCB P&P without heavy reliance on CV, instead trying to get everything aligned perfectly every time by pure mechanical brute force!
@@BreakingTaps if I understand your design right, that rotation stage will destroy all your beautiful Zabar xy-stage nm accuracy. Direct driving the rotation stage from the stepper will limit you to the (pretty bad) stepper bearing accuracy (compounded by the perpendicularity error of the shaft to the xy stage etc), the angle/step hold accuracy and repeatability, and create additional xy inaccuracy when you're not in the center of the wafer (which is after all the whole point of a litho stepper..). You may also see some fine scale mechanical vibration from the stepper/driver that will blur the image slightly in the polar coordinate system. You could try one of those cheap-ish stepper worm microscope rotation stages instead, with a conventional vacuum chuck on top (and just deal with vacuum hose manually, or program it to no spin around arbitrarily and tangle the hose).
It is COMPLETELY FASCINATING to watch you learn such a tremendous amount for all of us about what goes into the most rapidly-iterated technology ever with the most complex supply chain in history. Thank you!
This was really cool to see as a recently graduated EE. My senior design team built a maskless photolithography device and it was incredibly hard having basically no knowledge/experience working with optics. We had a very limited budget and even with extra funding we found that the limiting factor in alignment accuracy and pattern resolution was the optics portion. We managed to make our device for about $1000 with our smallest feature size at about 17 microns albeit we could not really make usable patterns. Definitely learned a lot during this project even if it was more than we could chew.
Dude, you should totally get into the kinematics of working on something like this. It’s very similar to what I work on 90% of the time and as I was watching you deal with the O-ring debacle of your third stage on the first machine, I was screaming at the TV “ oh man, I can fix that”. Lots of people struggle with mechanical alignment, but it’s really all not that hard if you know where to use the hammer and where to remove material.
Wow, so good! Really liked this video! The collimated light technique was super interesting. Looking forward to the engineering cut :D I know it might feel tough to have put in so much effort and only get 2 micron, but that is a win. The goal was what got you to where you are now, and it certainly appears like you're not far off. Big thanks from me for an amazing video!
This makes me giddy, good job :) i used to design maskless photolithography tools with the DMD chip. We used to use the mercury bulbs, but we had a very hard time transitioning to LED
I used to work with laser imaging systems 30 years ago which threw 10 micron spots onto large format printing plates. Green light really highlights dust particles, if you want to make sure your imaging target is dust free. : If the o-ring was just a dust seal, you could put down a ring of memory foam instead. : You can test how good your amber protector is for UV by using auto dimming sunglasses outside of the box. : You'll find that your red light and UV lights have different focal distances, so you may want to run a 'focus series' to find the ideal offset between the red-UV adjustment you make.: Keeping the center of the optics lined up helps you get a wider focused image. It's easy to be off center when it comes to lenses. again, green light and fog spray will help you visualize your beam. : Static electricity is really good at killing optical diodes. You'll have to wear a static strap to prevent this in the future. if you put your blown LED under a microscope, you may even see the fractured silicon inside if youre lucky : I think you did a good job considering that companies in the nineties would have thrown millions on their machines.
Not gonna lie, that was one of the smoothest transitions into a sponsor add, through the add, and back out that I have ever seen. I didn't even realize it was an add until over half-way through it.
At my work we make the wafer chucks for a few smaller and more specialized chip manufacturers, my job is to lap the chucks to a thickness, flatness, shape and roughness tolerance. Usually it’s around 300nm for roughing. Thought it would be cool to share, your channel is great and thanks for the good videos! 👍
Pretty amazing: I worked in aerospace back in the mid-70s and back then the big push from DARPA was to hit 1.25 um. Now you’re doing it in your personal workshop:-)
I look forward to the detailed long form video when it drops. I was already subscribed but updated my notifications so I don't miss it. Best and thanks for the awesome project.
8:50. Full time schematic and PCB designer here. You did great for your first time using solder paste. You needed a lot less, but you did great for a first try. After a few more tries you'll be a pro at this.
Awesome video! You’re a very clever man! Not only because of all the knowledge, experience and expertise you have, but also because of your deep understanding and the ability to explain things in a clear, simple, easy to follow along and interesting way! (If say from 1-10, 1 being a complete novice and 10 being an expert, I would say I’m around a 3) So even for someone like me, who only has a broad and basic understanding of how these things work, the way you made your video and explained things - I was able to follow along throughout the whole video, and because of that I could understand what you was talking about (not to a great depth mind especially for some parts) but because I didn’t get lost or confused at any point, I didn’t need to rewind to try and understand something you was saying or to try and understand how it worked… it meant I could really sit back and enjoy watching. Lots of UA-camrs have extensive knowledge, but most lack the ability to teach/convey it in a way that any person watching, can follow along and understand each part enough to enjoy watching it! Just shows your ability to teach others through engaging entertainment means you have a very deep understanding of what you’re talking about! So thank you! Have been watching your videos for a little while now, but luckily I’ve still got some videos left to enjoy ☺️ and from how you speak/make your videos, just by the thumbnail and title - I know if I’m going to enjoy watching it or love every single minute 😁 Lastly, I can’t wait to watch your follow up video explaining more about things you had to cut out from this video 😁 should be very interesting 👏🏼👏🏼
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
*Why bother with the rotation stage?*
Good question! It was for multi-layer alignments. I.e. pattern first layer wherever on the substrate, and then align all subsequent layers to registration marks on the first layer. In theory this could all be done in software (rotate the image digitally) but it introduces a lot of complexity to the software. I thought mechanical solution would be easier. But my design (direct drive to a stepper motor) was a flawed idea and in retrospect there are much better ways to do it.
The easy way to make a rotary vacuum too head is to use a dual hollow shaft stepper motor with a flexible silicone pipe at the bottom end and the tool head at the top. Obviously you have limited travel but you can easily do 360 degrees. That is how DIY pick and place machines work. No need for o-ring seals.
Honestly, when I first saw you assembling the rotation stage my first thought was "how will he ensure the rotational stage is planar?"
Based on your DOF with that level of magnification, I assumed that traditional indicators you'd see in a machine shop wouldn't be good enough to adjust it into focus across the whole stage. It's an unfortunate problem to run into, and besides having better methods of measuring to improve the accuracy you would either need some different equipment (surface/rotary grinders/etc) or a lot of patience and skill with sanding/lapping.
Regardless, I still loved the video and I hope you properly congratulate yourself for the wins you were actually able to achieve.
or you could locate the rotation by removing a part of the circular wafer, like it is done in industry. If you have a straight edge you should just need a couple of dowel pins to locate against. (in theory)
@@BreakingTaps to do layer-to-layer, you'll need to be able to figure out your pixel size in X and Y, as well as the skew of your optics. Hint: calculate your per pixel H move with X and Y, then do the same for V pixels. You'll have 4 calibrations numbers that you should be able to do all of the math with after. Also, you'll need to deal with your flat field correction for the objective. You can either shrink the exposure window or you can grayscale your output and correct with a digital mask. The easiest way is to use a white background and capture a blank window with the red wavelength with your inspection camera. You'll then need to map camera pixels to exposure pixels
What a time to be alive where one dude can recreate things that took a dozen of brilliant minds years and millions of dollars.
*hundreds of minds lol
it still took all those minds, and many more for the intermediate history that has made the task this attainable. It's really inspiring in so many perspectives
Like u doing math? Lmao
Sure, if you've never heard they found a way to put cheese in aerosol cans now.
We are standing on the shoulders of men, who are standing on the shoulders of men... dating back to the first dude who rubbed two sticks together and thought, "huh... Neat"
Greeting's from Intel's sub-nm research team! It's super incredible to see hobbyists getting to the point of making real semiconductors in the home shops.
I did my PhD on reducing resistance in chip-to-chip bonds, so if you ever get to that point where you want to make a really tiny sandwich, definitely get in touch down here!
Is that sub-nm (A7 process node)? What are some recent public things representative of the kind of work your team does?
The problems routinely solved in the semiconductor industry are incredible. This hobbyist level stuff is really impressive, but EUV lithography systems feel less and less real the more I find out about them. I'd love to have a few billion of your sandwiches :P
My understanding was that under 3 microns the silicon loses it's semiconductor's switching properties. When was this problem solved?
@@reddragonflyxx657 Sub-nm referes right now to any node from Intel ending in A, as those need Angstrom measurements. 20A/2nm is the first, followed by 18A, then likely 14A.
Most of the work from my team has been in the theoretical side of things, but you can see the first applications of our work in the Intel4 and Intel3 nodes. The 20A die of Arrow Lake will be the first with PowerVia, which will also mean it is the first designed with tools my team developed for sub-nm.
@@atomicrevenger4745 What are you on about? Silicon MOSFETS (for digital logic) with critical dimensions of ~3um are over 4 decades old.
If you meant 3nm, then it really won't matter anytime soon, since the upcoming "sub nanometre" nodes are still expected to have a minimum inter-transistor gate pitch of at least 30~50nm, that's *10~15x* bigger.
Also, while it's true that shrinking planar transistors (and even finfets) down below a gate pitch of around 20nm brings exponentially increasing static leakage and decreased gate control (think short-channel effects, like DIBL), silicon doesn't just magically "lose its semiconductor switching properties" at 3nm and in any case, we are unlikely to see any Fab actually get anywhere near that level of density during this decade.
The industry-wide switch to GAAFETs might, eventually, even allow for the introduction of commercially viable process nodes, with actual inter-transistor gate pitches in the 5~10nm range; however it's still far to early to say much with certainty about how GAAFETS will mature.
Very cool! First thought was anodic bonding, but I seem to recall that's only for Si-to-glass, not Si-to-Si?
"Screws take up physical space" is such a funny cad-to-IRL problem. I made a similar mistake with a lego robot arm I was designing in cad, and everything was very meticulously physically simulated to ensure all joints, cables, etc were perfectly fine.
And then I tried to build it and realized a few of my critical fastening pins along the arm's base were unable to be inserted because they would have to simply cease to exist and then magically re-manifest inside the already-assembled part, whoops.
I'm on a robotics team and this is why I cad all of my fasteners LOL
McMaster-Carr supplies 3D models for most of their fasteners, making it easy to import in fasteners so you don't need to model them yourself. If you have a lot it can definitely start eating up memory, so you can leave them hidden/suppressed for a lot of the design process, but having them in the model is really useful. McMaster also has add-ins for Solidworks and Fusion360 to directly search/import models instead of having to download from a browser then import.
¯\_(ツ)_/¯ I'm a Plumber, this is why we test things, because Humans are brilliant, but; also we bumble...
You could've made a lot of money if you simply solved that final issue of teleporting the pieces in place.
At least you learnt from your own experience here. As a welder fabricator I have PTSD about fully engineered designs that were physically impossible to weld lol. It would be upto to us to make modifications so it would work, and fully document those changes so the engineer could approve or revise them.
"I'm not an electrical engineer" - the guy who titled his video 'speedrunning 30yrs of semiconductor history'
LoL 😂.
Gotta hand it to this guy.
Although is there any level beyond semiconductors, for electrical engineers?
Somethin analog?
@@yasirrakhurrafat1142 I believe you can do a lot of hard stuff, high voltage for example can be challenging, because it's pretty deadly.
@@niklasbaack7238 might as well I guess.
With extreme caution.
Anybody can make anything with enough interest, time, research, and resources, but an EE can design something that your insurance company will let your employees operate. "I'm not an EE" is basically a line of distinction like someone saying they aren't giving medical, financial, or legal advice.
What’s the argument for it not being most important
i just realised something, in aerospace theres Aerobraking, and the cursed euphemism Lithobraking, which just means crashing into the ground, but if you use the terms the same way, Lithography literally means "Rock Writing"
we write patterns into rocks to make them think... literal runic carvings, we are magical rune carvers
My model rc planes did a lot of lithobraking
I have a friend who tells me that our whole profession is basically teaching rocks to electrocute themselves in ways we find pleasing.
And then the KSP term 'Lithostaging'
@@marsguyphil My favorite is a description of microcontrollers/processors as "just some rocks that we taught how to calculate"
If my memory is correct, it's called "lithography" because way back in the day (1800's), they used acid to etch patterns into limestone, and that patterned limestone could be used for as a printing block (basically, a big stamp) to repeatably print custom patterns onto paper.
Hey! Optical scientist here! I design and build these kind of setups for living (used to be custom microscope systems, but now space instruments). So fun to see what you can achieve with DIY stuff! Youre doing an excellent job!!
For me building and aligning is always the best part of a project. I think you can ease your alignment process a lot by using good alignment tools and mounts with the appropiate degree of freedom control. For instance, putting the objective on a z-translation cage mount would massively improve the alignment control. For the alignment tools I would suggest cheap ones screw on alignment targets and frosted glass disks. You can also make these yourself!
When I design and build systems I spend most of my time thinking how I will align it. Where is fine adjustment needed? What can be mechanically toleranced? Thinking ahead of all these things saves you massives amounts of time. I know for sure your next optical system will be even better! Keep it up😊
And maybe throwing away the 3D printer; every time I see a cantilevered 3D printed "support", I want to die...
@@Beregorn88Yes but waiting hours instead of days for delivery is such a win. 😋
@@MakeKasprzak that's only if the parts work: what usually happens is that you waste time and money, and then you have to buy the proper parts anyway. Or even worse, you go out of budget and can't complete the project at all...
Noted, thanks!
@@Beregorn88 That's not the fault of the 3D print though - that's the fault of the optomechanical designer. Using the wrong process for the job does not discredit the process itself. I absolutely think 3D printers have their merit in the prototyping process, also in the optical lab - you just need to know when it's the right tool for the job.
7:13 Neat! Haven't seen anyone shout out Huygens Optics channel before. I love it
Love his channel! He so much knowledge about optics, and does a great job explaining it. I'm always blown away by all of his demonstrations
@@BreakingTapsAgreed, I stumbled onto his channel a while back and was blown away at how good his explanations are.
@@BreakingTaps Yeah! His tiny glass telescope looks amazing.
Right?! Huygens Optics is like the equivalent of that lesser known restaurant all the best chefs love to go to after work. I see all my favorite youtubers in the comments over there! Super channel!
@@BreakingTapsas a litho engineer I’ve come to associate a Dutch accent with litho expertise 😂😂
It's truly stupendous of Zaber to have donated such a critical part.
It was amazingly generous of them!
Not really; they see it as an advertising opportunity and it's intrinsically a sponsorship; the product was even named.
So I work as an engineer developing lithography machines and it is funny to me how I have very little knowledge of the process as a whole, while noticing a lot of small things that would never fly at work. It is great to see someone build a working machine in an hour instead of focusing on a small gear in a very large machine for a year at a time.
I make parts for experiments that go into space and I don’t know what any of them actually do most of the time.
Do you have any advice for him?
Working on v2 so I'm all ears if you have tips! Top priorities are: a precision base for the stage (granite, optical breadboard, etc), reducing stack height of the stage to reduce abbe errors (and fix that stupid rotation stage), and re-arranging the optics so they aren't hanging/cantilevered or attached to the frame
It's like they say, no one person could design a modern computer mouse by themselves.
😂 what do you mean one machine? He built two machines in an hour!
41:00 Killing the LED is due to too high drive current. Look at the schematic and try to lower it or use different diodes that can handle it.
Or got killed by ESD, because those bare UV dies are really sensitive to it as well. Best to also add in the protective diodes that the manufacturer recommends, and which are almost never used. Shorted diodes either ran too hot, because of poor thermals, or got zapped by ESD.
@@SeanBZA There was a heatsink right next to the diodes - are those diode driven or from other nearby equipment? Might want to install temp sensors and/or increase heatsink capacity and/or airflow.
I think the main problem w.r.t the LEDs is at the 8:50 mark where the LEDs are being soldered. Note how they "soldered up a bit wonky", which might have seemed like a cosmetic issue for the most part at that moment. However, it has severe implications for the thermal performance of the system. Basically, the LEDs can't properly dissipate the heat they generate. As a result, their lifetime is shortened considerably. Running them at lower current will help some, but they'll still die prematurely. The real solution is (1) ensure proper dosing of the solder paste. There only needs to be just enough to flow the entire contact surface, but it shouldn't bead up, raising the LED away from the copper PCB. The other part (2) is to ensure that heat dissipation from the copper PCB is as efficient as possible. I'm going to assume that (2) is dealt with because apparently this sub-assembly was lifted from an existing/functioning machine. However, (1) still needs to be addressed.
I've dealt with this exact same problem with high-power SMD LEDs many times, so the above is my best guess based on 1st-hand experience. It doesn't discount other possible causes, but I'd really start tackling the problem here. It's a simple fix, in principle. That's the good news.
@@koraks9939 that's really interesting! Thanks for the write up
@@koraks9939 That indeed is probably the problem
You should have a chat with Cylo's Garage, he works on high precision mechanical systems down to the nm level.
Love his channel (and him!). Honestly hoping he doesn't see this video because he will be horrified 😂
@@BreakingTaps Oh come on, how could I hate on this!
@@cylosgarage 🥰🥰🥰
Do you know what kind of linear stage this video used? Or is the "accuracy" mostly from the encoders? @@cylosgarage
@@leonordin3052 More details on the Zaber page, but it's a crossed roller bearing stage. The absolute positioning accuracy and repeatability is in large part from the encoders, but the pitch/yaw/roll accuracy are all from mechanical design of the stage (i.e. encoders can't correct for a pitch error in the axis, that's purely mechanical/assembly)
(although since it comes with interferometry data you can compensate for pitch/yaw/roll errors in software, will touch on that in the next video)
The things you cut from this are my favorite parts - the really nerdy stuff. I care a lot less about watching the assembly of the frame, than I do about the design of the frame for instance.
I enjoyed this video, but I'm _really_ looking forwards to the next, regardless of the effort that goes into it.
Yeah sorry about that! The original plan was half assembly, half design discussion. But the decision to build a second machine wrecked my video-timeline-planning and it was hard to do both. At least this way I can spend gobs of time talking about stuff like abbe error compensation in the next video :)
Fun fact about standing waves in photolithography, they are worse the more narrow your source wavelength filters are. Standing waves are worse for laser and led sources vs hg lamps. Source: I make antireflective coatings for litho for a living.
Ah yeah, that makes sense! I assume a wider bandwidth will "smooth" the standing wave since the different frequencies all interfere at slightly different locations? I was impressed how crisp and "step like" the standing waves were! I had seen them in the literature before but never in my own tests. A quick post-exposure bake seemed to get rid of the worst standing waves luckily :)
@@BreakingTaps You have your work cut out for you! Early in my career I was also responsible for alignment and overlay matching a fleet of ASML tools. So many variables!
Putting a stepper motor on such a precise x-y stage kills all the precision to begin with
It was not the best of plans 😅
I was looking on in horror as he was assembling the tower of backlash.
@@Kenionatusme too, and thinking "you know, Thor Labs just... makes this kind of stage. You can just buy it."
I'm curious, wouldn't a structure like a 3d printer be better?
@@alexmipego it's not precise enough for this type of job
1 micrometer lithography. I am impressed. Really makes me wonder and appreciate how much effort and alignment goes into nanometer printing with lithography !!!
dude summoned the entire staff of ASML and their suppliers into his comment section
I've been working with a friend on a high resolution film scanner design for the past year, and I can tell you I powerfully empathize with all of those "I have to totally disassemble this and then reassemble and realign it moments and I would rather walk slowly into the ocean" moments. It's a real rollercoaster. BTW I'd spend the time to either figure out how those LEDs failed, or make them easier to replace. LEDs that failed once in a given installation are pretty much guaranteed to fail again.
Enjoying this greatly. I chuckled at the comment about professional use of 80/20. You might be surprised to learn how much 80/20 or Rexroth is used in production multimillion dollar tools by a variety of tool vendors...
and they cost multimillion dollars because they use 80/20. :D
Finally got a chance to finish watching this. I wanted to add that again, even on multimillion dollar production 300mm tools with the very nice precision motion components......there's still hundreds of hours spent during the life of the machine fiddling with 0.05mm shims.
Have you considered using a blu-ray pickup for both the UV laser and optics systems, and as a stage? A standard pickup has a nanometer accurate voice coil holding a lens, and a UV laser, as well as a photodiode, through you probably don't need that. I know there's been a paper on using them as a micro 3d resin printer. The idea is you use two pickups facing each other at right angles, since they only move in one horizontal axis as well as vertically, so you attach the sample to one pickup and use the other to scan the laser over the sample. I've used them to make a laser scanning microscope with extremely high magnification (diffraction-limited), so the potential is there. You can even use the photodiode to measure the distance to the sample with nanometer accuracy, which should help keep things in focus. The real limiting factor is that the pickup's coils, while extremely accurate, can't hold much weight. That means if you use one to hold the sample, you'd be limited to a couple grams max.
Fun fact: you can even use those same pickups to make an atomic force microscope
What are you some kinda wizard?
Wasn't there a youtube channel that did just that?
@@mr_gerber I think Dr.Volt (right name?) did one, mine's a separate project though, with
@@nilpybilpy4346 Yeah, that sounds right. Concept felt familiar.
Cool! Are you sharing your project somewhere public?
@@mr_gerber I tried posting the link from my phone, didn't go through so I'm trying on desktop (update, still didn't go through...). Here's the link for the project (just the project location relative to giithub's home page in case the issue was from posting a link): NilanEkanayake/LaserScanningMicroscope . It's currently not very polished, and will likely have bugs and require a decent amount of knowledge on the subject to build. It also uses a DVD laser (~650nm).
I mapped my z height using interferometry and DFTfringe and adjust my heights with a piezo linear actuator. X and Y were mapped using a reference mirror tilted at 45deg and about 12 feet to calculate any wobble (it showed that i had a slight twist and bow that i compensated for in software). I also dont use a lens on my camera since the light coming out of the objective is suppose to be parallel. A cheap photoresist that i use is D&S Posicoat used for metal etching. I can easily get ~800nm features with it (its as far as i could test before my daughter was born). I took alot of inspiration from hyugens optics with a mix from applied science. I have mostly made mesfets due to their simplicity. Also materials project is a great resource when trying to determine what materials to use, bandgaps, mobilities, lattices, energy levels, and compatible substrates.
Seconded, while troubleshooting uniformity I've unplugged the piezo signal on a maskless aligner during a write. Work it's able to do to keep focus cannot be understated.
Can I pick your brain about the camera comment? I have a secondhand stereo microscope from Vision and am trying to mount a camera on one of the two optics paths. I believe the light is parallel in these stacks but a camera in the path doesn't work right... How do I reason about camera sensors with parallel light beams? I thought we had to project a focused image onto the sensor?
would love to see more! let's go for doom on a self-made chip? Or is that crazy
Wikipedia says the 386 processor was made on a 1 micron process node... so maybe?
RISC-V could be a thing here...
Doom on FPGA has been around for a few years now. Let's see it on raw silicon
@@andrewferguson6901 a doom asic would be pretty cool tbh.
Hrmm. I would have expected the construction of a 1um machine to look like, you know, a high end machine tool. Lapped surfaces, air bearings, granite, that kind of stuff. Machined aluminum on a warped piece of sheet metal with a bunch of shims, holy wow, getting what you did is super impressive!
I‘m not even halfway through the video and I’m already impressed
It was kinda funny watching you build a 2M$ machine, yet struggling to make a simple metal box :D But I also saw that you improved massively just after the first try, so it shows how smart you are and how quickly you learn.
Amazing job with the project, I hope we'll see more of it in the future!
Haha yeah, the irony wasn't lost on me either 😂 Lessons were learned!
This is something you can watch twice and still learn something. My heart goes out to you. One component can really drive you nuts if it is not easy to access, especially when you have to re align the optics. I am looking extremely forward to your next video 👍
The machine at 7:06 looks nice ; ) Cool video, looking forward to the next one!
And is cool to use! We got that very uMLA last year and oh boy it sped up my lithography experiments from one layer per day to 2-3 x3 Ability to expose in Gray-scale is fun too
Heidelberg reminds me of working in my print shop 40 years ago...
Can we please all take a moment to appreciate the section (15:10 after assembly) where you explain everything you would've designed differently. There is so much to learn from there!
This is possibly my favorite channel. You have the perfect personality for doing all the things you do and talking about them, no idea how you are able to stay so optimistic and calm when I know how darn frustrating and tedious many of the issues that can and do come up are. This particular video came out at a really great time for me as I'm about to be working on literally the exact same thing (something I've been working towards for years but finally now have the time)
You might also like Dalibor Farny - he makes Nixie tubes and shows a lot of the "I tried this and..."
Loved the edit at about 26:30. I absolutely love everything you've been doing. Its wild to see just how possible it is to recreate this kind of technology in a home lab.
This is really amazing. Keep up the awesome work! I really appreciate that you show both the positive outcomes and what went wrong. The failures, setbacks, and frustrations are what I relate to the most. Most creators only showcase their progress, but seeing someone question their life choices, just like I do when working on a project (though not as complex as yours), really motivates me to keep going. So, thank you!
Thank you for the furlong conversion. Helped a lot!
My really old university professor (and add 50 years) used to love to require final answers in furlongs or, furlongs per fortnight when applicable.
For the UV LEDs, you might want to check the paste you used, if it is WS type and you didn't ultrasonic it or flush it with water and often IPA after that, there's a good chance the acidic PH ate away at everything beneath them and turned it into a conductive mess. WS=water soluble, NC=no clean. WS is good because you CAN clean it up readily and remove flux residues, NC is okay for digital stuff etc, where a little bit of flux isn't too big a deal even if it can cause megaohm parasitic impedances. BUT I'm not suggesting you'd want to use NC, as the hot environment might also be enough to activate it enough to slowly attack the LEDs while they are on if they get hot enough, ideally you'd use WS and clean it off so no flux is left, it's just that WS will simply corrode it with time, and not just while it's hot.
Else heat indeed fused them shorted, suggest monitoring their temperature as part of the process or upgrade heatsink and cooling.
Oooh, I didn't realize that was a thing. Just checked and it's "kester ep256 no clean", also very expired (got it a few years ago for a project that I never completed). Good to know! I'll pick up some WS for the next batch of LEDs.
7:40 there are at least 3 universities in the US with the abbreviation CMU, it is probably wise to specify which one you mean.
in this case, its Carnegie Melon University, for those curious.
Whoops, I had no idea. TIL!
Congrats on the mammoth achievement! I yearn for the day when custom IC design becomes as easily accessible as PCB printing - a true superpower in everybody's hand. Surely, a step in that direction. Amazing work. Keep it up.
Cool stuff, I attempted something similar but the deposition of metals was too expensive for me so I stopped at the lithography part. Some advice,
Mount everything to a granite surface plate, you can use a diamond brazed bit and a router to bore holes into the surface plate then glue inserts to secure any apparatus. This makes aligning things much easier.
Not all DMD offered by TI are the same, the one you used has really poor optical performance when it come to
Good stuff man
What do you mean you are not an engineer. You are a very talented engineer. Ability to pull this off and passion is very rare. Consider me ur new fan and a subscriber. I work in this industry and this video was fun to watch
That's what I was hoping to see in a UA-cam video one day.
I love that this is possible in your garage for a reasonable price
Very cool that you mention all the mistakes as well as the victories, for learning
It just hit me that a DMD chip is basically a really, really small flip dot display. In a way
It's also a modern mechanical television.
I appreciate the video time you spend on the design review (e.g. socket heads colliding with the cover).
Such a cool project! This is the kind of thing I've often imagined doing, but doubt I'll ever get to trying.
I want a video of you partnering with Applied Science. You both absolutely stun me with your expertise and video qualities for such complex topics. I learn so much every time. Amazing work!
Just a tip: The "nose" on a stepper motor is there to align the axis where you planned. This takes away any tolerance in the holes/bolts.
If you had matched the hole for alignment you could have enlarged all the bolt-holes and still had motor centered on fixture as planned. Not a big deal in this design but would have been important in a design where you rely on shaft position for gear mesh or belts etc.
TIL, cheers!
this really is awesome, i have worked on Nikon Photolithography tools and now electron beam lithography tools, this really hits the basics very well. This guy is next level smart, time for him to move into semiconductor land.
The dog at 19:55 is like "cool man but what about my food?"
Confused dog cameo had me coming to the comments looking for the mention of it.
Makes one appreciate the 3nm process all the more. Well done, the best bits are the mistakes for people to learn from, thanks for the candid approach to the videos.
Wow what a journey! I suggest trying a dedicated driver for the LED's. They probably died do to high current and/or heat.
Perhaps this is why people prefer working in the digital space. Physics is so much trouble.
But I’m really happy to see your knowledge and determination on display. Stretching is edifying.
Notes on the diode test setting of the multimeter for LEDs: Standard silicon diodes will read around 0.65V, red LEDs will be closer to 2V, and blue will be closer to 3.4V. Larger voltages for shorter wavelengths, in general. There's a lot of variation between different LEDs, so treat these as rough numbers. When you measured the red LED board one direction and saw 0.66V, that was probably a reverse protection diode. I couldn't find the LED PCBs from TI's docs to check, but the Hacker-Fab board has two LEDs in series on theirs. That means that the forward direction for the red LED board should read around 4V. Depending on the multimeter, 4V is too high to be read by the diode test function, so it'll show up as overrange. The 0.071V you measured for the UV LEDs is pretty close to a short. They are definitely toast (but shorted, not open circuit). If you wanted to measure the 4V of two red LEDs in series with your meter, you'd maybe want to look for one that has a zener diode test function as that'll read up to a much higher voltage.
(And I'm really looking forward to the 2nd video with the engineering deep dive.)
This is way,way beyond my reasoning and ability to logically form a conclusion to this technology.I applaud these brilliant entrepreneurs! Hopefully their efforts are rewarded and not stolen.❤
It's amazing to see that you are as bad as I am at designing mechanical parts. It gives me hope that the next step, doing something amazing, is right around the corner.
Haha yeah, it's all fun and games until you have to assemble it 💀
@@BreakingTaps Cursing the idiot who designed this latest puzzle last month is great motivation to become a better designer.
You're really smart, we need more well not people but brains like yours in the world.
Holy crap I saw this video recommended and was like "This is legit gonna be such an awesome video" and then realized it's from my favorite creator. So cool!
D'awwww 🥰🥰🥰
You're accomplishing things I could only dream of. Doing obscure but extremely interesting things the hard way used to be my hobby. I did some truly interesting things but... with age came an essential, though benign, tremor. Looking forward to the next video. I immensely enjoy watching them.
After taking a materials engineering course, I understand just enough that I can follow your videos, but you still constantly teach me and expand my knowledge, keep up the awesome videos man!
Thermal expansion will rock the focus every time.
I really can't thank you enough for showing all the annoying little mistakes, like e.g. forgetting that screws actually take up physical space.
When I make similar mistakes I often start seriously doubting my higher brain functions. Seeing someone admirable like you, fall into those same pits, really starts to convince me, these mistakes are essentially unavoidable.
I wish you the best of luck in your further endeavors from Germany!
Recreating lost technology is immensely difficult! It the little things (seldom documented) that are the most challenging to rediscover, things that no one at the time considered important. Keep at it you will figure it out!
Must admit, the fact that the whole "calibrating texas electronics lenses" section was filmed out of focus was way funnier than it should be. I understand having to put all the focusing mana into the primary objective though, great video
That zaber unit is 🤌
It's soooo good. 🤩 Legitimately one of the nicest pieces of equipment I've worked with. Excited to improve the machine in version 2 so that we can really push the limits of the stage.
@BreakingTaps maby try using hydrofluroic 1-2 percent to clean before addition of the photoresist for better adhesion. The age of the photoresist and how it is stored will impact the quality. I would suggest a photoresist that is newer
Shout out to zaber for just donating it. I just took a look at the pricing for those things and that is definitely not the kind of thing you just casually pick up for a project lol. Very kind of them to just donate it without the usual quid pro quo promotional stuff.
So fun to see them mentioned in the wild. I do biomedical and a zaber stage is on my wishlist for a custom microscope.
Yeah, until I saw that I thought this was gonna be a practical How-to DIY video.
A tip for metal work from a welder (if you do more metal work): An angle grinder with a flap disc. Flap discs are like what sandpaper is for wood, but for metal. They are great for rounding/cleaning edges and getting mill scale off. It makes welding a breeze and would have taken less than 5 minutes to clean all the edges of your pieces.
When I worked at a mom and pop fab shop doing odd jobs like store racks, we would use a lot of in-house cut sheet metal and we had boxes of the flap discs for edge cleaning simply because while they are quite consumable, they work VERY well.
Such vast latitude in DIY and hobbyist opportunities & capabilities is what makes a super power, not its nuclear arms stocks.
Impressive job. Top notch patience and dedication.
Kudos! from Kenya.
A better option vs your steel sheet metal plates would be to use waterjet and machined MIC6 aluminum plates, at least for anything that can be subsequently aligned with gauge blocks.
So awesome that you built this by the way :)
I remember learning all those same lessons as a junior mechanical engineer.
Also hilarious how you ended up back at building masks 😂😂😂
Stop painting ! You are adding errors. Dude you are amazing. Precision manufacturing and clean environment are the basics.
would be really cool to see a like 80s era transistor, opamp, whatever like all the way, as a series. so getting the wafer exposure done enough, maybe place transistors for some existing design, figure out bond wires, make packages (did they stamp the cans out?), and putting it in a basic circuit to do something
Informative, innovative, creative, and entertaining 👌 👏 👍
🥰
@@BreakingTaps 1-2 um is impressive. Watching you create solutions is inspiring. I'm sure you'll continue to improve upon your results.
@@BreakingTaps Have you explored Reactive Ion Etching?
@BreakingTaps Forgive me I see you did a short video on Ion Etching. 👍
The 'prototyping' with 3d plastic before going to metal makes a lot of sense. The improvement between the two seems huge. Awesome work and video!
0:01 OK now I see why the logo is a triangle!
Decades back, my grandfather had a small wood shop setup in a shed. Now someone who seems to be roughly the same age as me, essentially has a cpu shop. This is equal parts awesome and amusing.
Why the rotary stage?
generally for such application all the components should be precisely grounded,using indicator during installation checking parallelism and straightness .xy stage should be bolted on granite plate, the whole setting a vibration isolation stage
Use it to build a PC for scratch
Thank you very much for this video!
I happen to work on the first in field GCA steppers. Got to help design the FAB enclosures, Fab placement and involved with the installs up until we put them into full operation mode at which point I moved on to other projects. Worked with field service techs and engineers while we tried to get them on line. Good time!!
Watching this time flew; it felt like 10 minutes. Great video, as always. What chip are you planning on making with those machines?
Thanks! I want to make some image sensors (either CCDs, or the very early one-transistor CMOS sensors)! Just really neat devices and it'd be fun to record some video on a DIY image sensor. After that I kinda want to play with analog stuff, maybe some neuromorphic chips that rely on analog values etc. Basically looking at interesting devices that don't mind having kinda big feature sizes :)
It’s cool to see you using a zaber stage. I had a couple internships there during college and did testing on their first generation linear motor devices. Crazy how accurate and fast they can be.
You should *really* have read the TI documentation regarding DMDs used in the UV range, as well as that regarding your chosen chip.
To cut a long story very short, you will definitely ruin (or at least degrade) that DMD if you blast it with < 400nm light. Furthermore, the optical properties at UV wavelengths of components meant for use in the visible spectrum can cause *lots* of issues, such as:
- A lot of light being reflected back into the LEDs,
- UV absorbtion causing extreme heating of stacked components without good cooling. The glass covering the DMD and any of the optical lenses where thermal expansion would cause a large deviation in the focus or worsening of optical abberations at UV wavelengths. Even the metallization of the DMD mirrors in those consumer-grade chips can sometimes absorb a lot of UV Iight.
- Ineffective anti-reflection coatings.
Also make sure to check the dispersion of your optics, or even better, use the lowest power light source with a wavelength that is as close as you can get to that of your exposure LEDs (without noticably exposing the resist), to minimize focus errors caused by dispersion. I can't think of a single good use for red light on this kind of system, so maybe use that channel for a 450~460nm low-powered illumination driver.
You also need to ensure that the blue (>400nm) light that hits the DMD is properly diffused and collimated (I'm unsure how the optical train of that dev kit is configured, but you definitely can't just shine those LEDs directly onto the DMD.) TI has system design examples for each DMD that they sell to help you with this. Uneven input lighting doesn't just translate to a different dose-rate at the wafer, it also causes uneven heating and expansion of the DMD's mirrors. You could even be locally exceeding the specified absolute maximum spectral illumination intensity, thereby constantly degrading the reflectivity and/or the dynamic performance of the mirrors or the underlying CMOS latch at those hotspots, which would eventually lead to either 'dark' spots or 'stuck' pixels.
I didn't see how you were driving the LEDs, but you can't just re-use the integrated drivers and need to use a good quality, low noise *linear* current regulator with the usual UV, OV, OL and transient protection features to protect the LEDs and to keep them at a constant brightness. You should ideally take great efforts to keep the LEDs at as perfectly constant of a temperature as you can manage (also keep them under 50°C, most ideally around 30°, and only drive them at ~half the recommended current/power if you want them to last for a decent amount of time.), mainly to maximize the amount of time that you can use a constant exposure time.
It's no fun at all to constantly have to re-run your exposure optimization process, just because your exposure intensity keeps randomly decreasing as the components age. Or, you know, invest in a closed-loop exposure dose control system to automatically compensate for the average brightness variations at the wafer (which will only work if each LED ages at nearly the same rate over the entire projected field, but I digress).
Finally, I know almost nothing about the mechanical side of engineering, but I do know that NEMA motor has no right to be directly coupled to the image plane.
You should also completely discard that magnetic attachment system. Those magnets aren't going to snap back within 10um of their last position. It would be *much* better to lap the chuck against its mating surface, or by some other means generate a large, flat and even contact surface for it to repeatably and mechanically mount against.
Cheers! Some of these are definitely on my radar, just didn't have time to fully implement. Taking notes about some of the other points you brought up and will try to incorporate them in future iterations 👍
@@BreakingTaps Thanks for taking the time to read all that *and* for writing a reply! It is much appreciated. (PS. Please excuse my bad english. I do try my best, but languages have never been something that I could really get the hang of.)
I would really love to see this project succeed so, if you ever think you may have any questions that an EE might be able to help with, don't hesitate to contact me.
I even have some excess free time at the moment so I wouldn't mind doing some quick circuit design/simulation and/or PCB design and layout work for this project, especially if it's regarding something that I'm experienced with (obviously, I wouldn't have any IP rights to the designs and don't want to be publicly credited or thanked at all, nor would you be obliged to use any of my contributions. I just want to help you to avoid some of the common pitfalls and "traps for new players", that I'm familiar with.)
you have an amazing editing style. often times these techy/sciencey videos can get pretty boring to look at, but your 46min video was over in a heartbeat. earned a new subscriber! keep up the great work 🫶🏼
15:44 Insert "This Old Tony" joke here
33:50 where you have the lenses lying on their optical surfaces is giving heartache xD
Great video as always :)
hello again!
That's one of the most "real" and dignified sponsor promotions I've seen - congrats - you're doing so much right with your content. Design for maintainability next ;)
Why even have the rotation stage on the maskless system? Seems like a ton of complexity (and stacking errors...) for zero benefit when the projection can be rotated in software.
In theory it could be handled purely in software, but makes the software notably more tricky. With mostly mechanical alignment the optical/software alignment is simple. "find feature, fix minor rotation error, display pre-cropped image". But if you want to allow arbitrary rotation you need to identify where you are, translate from mechanical space to image space, rotate and crop out the relevant portion from a huge pattern (my test images were like 50,000 x 50,000 pixels before getting cut up for stitching), display and hope all your calibration is dialed.
Totally doable, just a different set of challenges. I gambled that the mechanical side would be easier/faster :) That said, v2 is definitely going to have a different design. I made some silly choices here :)
@@BreakingTaps Yeah, I guess I'm approaching this from the point of view of mainly a software engineer :)
I will likely be working on maskless photography soon-ish, and my thoughts are that since the 70s/80s two technologies have really moved forward: the existence of DMD/DLP (which you covered) and availability of compute power/algorithms. Meanwhile, machining precision elements is still expensive. And you see reliance on software solutions in manufacturing elsewhere, too - imagine modern PCB P&P without heavy reliance on CV, instead trying to get everything aligned perfectly every time by pure mechanical brute force!
@@BreakingTaps if I understand your design right, that rotation stage will destroy all your beautiful Zabar xy-stage nm accuracy. Direct driving the rotation stage from the stepper will limit you to the (pretty bad) stepper bearing accuracy (compounded by the perpendicularity error of the shaft to the xy stage etc), the angle/step hold accuracy and repeatability, and create additional xy inaccuracy when you're not in the center of the wafer (which is after all the whole point of a litho stepper..). You may also see some fine scale mechanical vibration from the stepper/driver that will blur the image slightly in the polar coordinate system. You could try one of those cheap-ish stepper worm microscope rotation stages instead, with a conventional vacuum chuck on top (and just deal with vacuum hose manually, or program it to no spin around arbitrarily and tangle the hose).
@@hugov392 Agree, was going to say that microstepping at this kind of resolution is going to be an issue.
It is COMPLETELY FASCINATING to watch you learn such a tremendous amount for all of us about what goes into the most rapidly-iterated technology ever with the most complex supply chain in history. Thank you!
0:32 "last 30 years of progress", you mean last 50 years
70
This was really cool to see as a recently graduated EE. My senior design team built a maskless photolithography device and it was incredibly hard having basically no knowledge/experience working with optics. We had a very limited budget and even with extra funding we found that the limiting factor in alignment accuracy and pattern resolution was the optics portion. We managed to make our device for about $1000 with our smallest feature size at about 17 microns albeit we could not really make usable patterns. Definitely learned a lot during this project even if it was more than we could chew.
Dude, you should totally get into the kinematics of working on something like this. It’s very similar to what I work on 90% of the time and as I was watching you deal with the O-ring debacle of your third stage on the first machine, I was screaming at the TV “ oh man, I can fix that”. Lots of people struggle with mechanical alignment, but it’s really all not that hard if you know where to use the hammer and where to remove material.
Man, you take on the hard projects and your tenacity in completing these project is impressive. Really appreciate all you do!!
Wow, so good! Really liked this video! The collimated light technique was super interesting. Looking forward to the engineering cut :D I know it might feel tough to have put in so much effort and only get 2 micron, but that is a win. The goal was what got you to where you are now, and it certainly appears like you're not far off. Big thanks from me for an amazing video!
This makes me giddy, good job :) i used to design maskless photolithography tools with the DMD chip. We used to use the mercury bulbs, but we had a very hard time transitioning to LED
I used to work with laser imaging systems 30 years ago which threw 10 micron spots onto large format printing plates. Green light really highlights dust particles, if you want to make sure your imaging target is dust free. : If the o-ring was just a dust seal, you could put down a ring of memory foam instead. : You can test how good your amber protector is for UV by using auto dimming sunglasses outside of the box. : You'll find that your red light and UV lights have different focal distances, so you may want to run a 'focus series' to find the ideal offset between the red-UV adjustment you make.: Keeping the center of the optics lined up helps you get a wider focused image. It's easy to be off center when it comes to lenses. again, green light and fog spray will help you visualize your beam. : Static electricity is really good at killing optical diodes. You'll have to wear a static strap to prevent this in the future. if you put your blown LED under a microscope, you may even see the fractured silicon inside if youre lucky : I think you did a good job considering that companies in the nineties would have thrown millions on their machines.
Wow
Man your video is AWESOME! please consider making it by parts, releasing one each week, this way you will get people engage and anxious about it!
Not gonna lie, that was one of the smoothest transitions into a sponsor add, through the add, and back out that I have ever seen. I didn't even realize it was an add until over half-way through it.
At my work we make the wafer chucks for a few smaller and more specialized chip manufacturers, my job is to lap the chucks to a thickness, flatness, shape and roughness tolerance. Usually it’s around 300nm for roughing. Thought it would be cool to share, your channel is great and thanks for the good videos! 👍
Pretty amazing: I worked in aerospace back in the mid-70s and back then the big push from DARPA was to hit 1.25 um. Now you’re doing it in your personal workshop:-)
I look forward to the detailed long form video when it drops. I was already subscribed but updated my notifications so I don't miss it. Best and thanks for the awesome project.
I'm really impressed what you have done. Those kind of tasks was done by whole teams not a single person.
This is so cool. Rediscovering that repairability is something we should value more.
8:50. Full time schematic and PCB designer here. You did great for your first time using solder paste. You needed a lot less, but you did great for a first try. After a few more tries you'll be a pro at this.
Awesome video! You’re a very clever man! Not only because of all the knowledge, experience and expertise you have, but also because of your deep understanding and the ability to explain things in a clear, simple, easy to follow along and interesting way!
(If say from 1-10, 1 being a complete novice and 10 being an expert, I would say I’m around a 3)
So even for someone like me, who only has a broad and basic understanding of how these things work, the way you made your video and explained things - I was able to follow along throughout the whole video, and because of that I could understand what you was talking about (not to a great depth mind especially for some parts) but because I didn’t get lost or confused at any point, I didn’t need to rewind to try and understand something you was saying or to try and understand how it worked… it meant I could really sit back and enjoy watching.
Lots of UA-camrs have extensive knowledge, but most lack the ability to teach/convey it in a way that any person watching, can follow along and understand each part enough to enjoy watching it!
Just shows your ability to teach others through engaging entertainment means you have a very deep understanding of what you’re talking about!
So thank you! Have been watching your videos for a little while now, but luckily I’ve still got some videos left to enjoy ☺️ and from how you speak/make your videos, just by the thumbnail and title - I know if I’m going to enjoy watching it or love every single minute 😁
Lastly, I can’t wait to watch your follow up video explaining more about things you had to cut out from this video 😁 should be very interesting 👏🏼👏🏼