Damn! Your videos are so very interesting that I could not tear myself from RUclips until 5:oo in the morning! You need to include a warning about how addictive your videos are!
Great stuff, really impressive machine! I agree with the leadscrew analysis (essentially the same periodic wormgear behavior that telescopes calibrate out), but wanted to throw out another potential contributor: motor cogging. E.g. the stator slots prefer to settle into low-energy positions relative to the permanent magnets and experience a force when moving in/out of those positions. Nicer servos will use tricks like skewing the magnets to reduce cogging (or ironless motors which have no cogging, but also very poor torque). It can also be compensated in software by building a "cogging map" which commands the motor to use more or less current at specific locations in the revolution so that cogging is nullified. The periodicity depends on the number of slots/magnets but will be higher than just once-per-rev, so is probably a higher-frequency signal overlaid the 4000-step non-linearity chart you showed.
You deserve a pat on the back, even real steppers suffer from many on the same issues and have a variety of software calibration files to account for this.
Very impressive! I admire how far you got in such a short time! When I worked for ASML in 1998-99 as a software engineer, I learned that there are some very weird problems that most people might not think about. The machines i worked on had 23 degrees of freedom, moving a 30kg marble slab with a wafer around at up to 6m/s or so. Of course the reticle had to move in opposite direction (but at 1/4th the speed IIRC) and there were counter-weights to keep the machine in place. The motion control of the linear actuators needed to compensate for "cogging" (i.e. variation in the magnetic field depending on where the magnets are), and things like the bending and un-bending of the electric cables that are attached to the stage. They used a Hewlett-Packard laser interferometer to measure the position of the wafer stage but I think that was just for rough estimates: the wave length of a red laser is much longer than the accuracy of the machine (180nm in that time). The software was responsible for basically keeping the speed constant, so it gives the motor a big push, then lets the stages "coast" to the other side with just a little bit of compensation for irregularities along the way. Then at the end of the track the motor would be pushed in opposite direction to stop the mass from moving. And of course it's "fun" if something fails and the pushback never happens; when a wafer stage crashes, it will cause significant damage that will cost a lot of time and money to repair. I remember someone telling me that in a fab, even opening a side panel causes so much disturbance to the air flow that it took an entire day for things to settle and make the machine usable and productive again.
@@barmalini I just started as a development engineer at ASML's German optics supplier and am continually amazed at the engineering they do (although our stuff isn't too bad either).
I worked in Texas Instruments Optics Department in Dallas Tx, where we perfected the coating for the top lens of the digital light chip. After our Optics department made the coating repeatable, TI moved the coating to SE Asia.
6:25 - I want to mention that "Trinamics" driver chips produces no noise and is pure analog signal to drive the servo/stepper motors, really good for this micro stepping to remove any unwanted vibrations.
Oh no, Trinamic stepper driver ICs still make lots of noise, but humans can't hear their ultrasonic PWM frequency. Yes, they're great chips, and yes their 35kHz (configurable) hard-switched PWM is adequate for most purposes, but sadly no they're definitely not pure analog and definitely not vibration-free in general. If you can find, design, or hack a PCB to use an 8MHz external clock for the stepper driver IC, then you can configure it to an audible 15.6kHz chopper frequency and listen with your own ears. I definitely don't work for Trinamic and I only have direct experience designing around the TMC2209 and TMC5160 ICs. If you've found a Trinamic IC that is truly analog, or can generate >100kHz PWM to allow efficient external filtering (without overclocking outside maximum ratings), I'd love to hear about it.
@@W77W well I would use trinamics for this project since it eliminates noises compared to the drivers he is using at the moment. But I’m not sure if ultrasonic noise makes no difference compared to the lower frequency noise which humans can hear. Trinamics are not pure analog but they are very close to it. I can’t think of any better driver chips other than trinamics. Another problem is that I would not use stepper motor, they are no good for 1 micron precision. Servo motors with close loop is better for precision positioning and I would also include an optical encoder with incremental accuracy of 1 micron. Trinamics TMC4671 can switch at 100khz. Here is the link: www.trinamic.com/company/news/news-detail/implement-your-servo-controller-in-a-day-with-the-tmc4671/
@@ShopperPlug Thanks for the quick reply and thanks for the link, I agree with you 100% that the Trinamics are the best single-chip solutions available (to my knowledge). I just wanted to throw in my 2 cents of caution because I did have an issue with a PCB drilling machine, where the 0.1mm micro-drill bits were breaking because the XY stage was vibrating ultrasonically enough to flex the drill bit, even when all axes and spindle were "stationary".
@@W77W Wow 😮, that is really interesting, how did you come up with the determination of the fact that the Trinamic’s ultrasonic switching is what caused the 0.1mm drill bits to flex and break? Would like to see your project if you documented it, I have plans to build a precision drill machine for PCB as well. Right now I’m gathering plans and materials for building a precision sub-micron laser PCB photoresist sensitizer. I’m planing to use custom built linear induction motor, no servo or stepper motor. I will use heavy triple-A above laboratory grade precision granite for the base. I’m also considering to use air bearings. This is the basic design I’m planning to use for the CNC laser sensitizing system, it is used by all high precision companies that provides nanometer positioning accuracies, and ultrasonic vibrations does not affect the x/y system since it is 100% frictionless (not considering the air molecules) when driving the linear induction motors: static.pi-usa.us/fileadmin/_processed_/7/d/csm_Air-Bearing-Planar-Stage_A-322_400w_ed5e39796f.jpg
@@ShopperPlug I'm really sorry but it's not my machine, so I don't get to share the build. I can say that epoxy granite (en.wikipedia.org/wiki/Epoxy_granite www.adambender.info/post/2017/03/25/epoxy-granite-machine-frame-how-to) is amazing, superior to the natural material in many respects. Being able to pre-cast notches, rails, and screw threads is really nice. Of course you need something flat to start with, but you can cast your epoxy granite upside-down on your surface plate to transfer almost all the flatness properties from one to the other. In our case we first stuck a precision parallel slightly above the surface plate, then made the casting to generate a precisely parallel-sided rectangular slot in the base, connected to the surface by an imprecise window, so it's roughly an inverted T-shaped cross-section like a milling machine table's T slots. I don't know if the thermal expansion properties are good enough for nanometer precision, but they're certainly good enough for a drilling machine. Just don't cast your very expensive parallel into a block of epoxy granite without a solid plan to get it back out. In our case we just chilled it to shrink the steel. It's also surprisingly easy to make DIY air bearings (ruclips.net/video/K_N_h_mKf-4/видео.html). Machining graphite is trivial, so you can easily make exotic bearing configurations. For example, you could machine a roughly X-shaped piece of graphite very slightly oversize for the parallel-sided T-slot, squeeze the X so it flexes very slightly, and wedge it into the slot (rather like installing pistons in an engine). Now you have an air bearing axis that's as precise as the parallel you used to cast the slot, and it's fully dimensionally constrained by the 4 sides of the slot, at least up to the elastic stiffness of graphite. You just mount the next axis directly to the graphite block. I suppose if you need a really large travel you might want two parallel slots so you don't have to buy/make a huge precision parallel or a huge chunk of graphite, but the idea is the same. Again I'm not certain you'll get nanometer precision out of this, but for a drilling machine it's great. If you can't follow my description I can draw up a sketch. I connected the dots to the ultrasonic resonance by several rounds of troubleshooting. I started by observing that on smaller PCBs (~5*10mm) the drill bit would snap very regularly, often at the same location on the PCB. Larger PCBs were immune. I figured that something must be moving, but I had no idea if the spindle had weirdly inconsistent runout, or the stage wasn't stiff enough, or the PCB was flexing, or maybe thermal expansion or air/vacuum pressure variations. I made several wrong guesses, then eventually hooked up a cheap continuity probe from the tool to the PCB to see if there was low-speed drift (thermal expansion driven). I use this probe all the time on my milling machine, so I was really confused when instead of a sharp on/off at the edge of the PCB, the LED would slowly get brighter as I bumped into the edge. Again some more wrong guesses, but I eventually hooked up an oscilloscope to the continuity probe, and clear-as-day you could see the 23kHz vibration on the 'scope. There's no way 23kHz was coming from outside the machine, so I pretty quickly followed it back to the source. Dialing up the PWM frequency and decreasing the holding torque solved the problem, but a heavier stage would also have helped.
I tried that and doesn't work very well. After that you also need to pass the image through the microscope objective. You add more and more lens. I got the best results in my test with Huygens metod 👌
Not sure how one of your lens grinding appeared in my feed but what you are dong is fascinating i'm so intrigued by the level of detail you have gone to in each stage of this project is truly masterful. I was considered building a homemade telescope and was stunned by the amount of work that goes into hand grinding a lens. Your solutions are well researched and executed and captivating to watch. Thank you for shearing.
Thank you RUclips for showing me your fascinating channel. I particularly enjoy watching people build equipment that would cost thousands and thousands out of various bits and pieces. You are doing this in a field that is both rare on RUclips and of particular interest to me. Thanks!
Wow. I have mechanical noise from much larger leadscrews when 3D printing- and I only am dealing with movements around 1/10th of a millimeter. This kind of stuff seems orders of magnitude more difficult. This is my favorite project so far. Amazing. Keep up the good work.
I used to work on the product team that made exactly what you need to get this working! We made twin axis heterodyne interferometers for these applications. Don't seem to be many of them floating around second hand though!
Hi Jeroen, I'm so glad that I found your channel. Also, I just wanted to let you know that diffraction is a classical phenomenon, not a quantum mechanical one (3:19). Thank you for making these videos!
@@HuygensOptics Thank you for your reply. It seems though having merely the wave property does not make it a quantum phenomenon. In fact, nothing in the theory of diffraction states that light is quantized, which is another necessary property for a quantum phenomenon. Maybe if you had a single photon emitter and you observed interference then you can say it's a quantum phenomenon. If you look at it as a Venn diagram, geometrical optics is a subset of wave optics (diffraction), which is a subset of quantum optics. In reality, quantum optical phenomena such as optical tweezers and Rabi Oscillations are much more difficult to observe. I'm sure you understand the matter very well, however I just don't want the viewers to get confused.
Very interesting. Thank you! You did not talk about influence of temperature on the device dimensions. Do you consider it as a problem? Aluminium expands ca. 0.1 percent over 50 oC. How do you make sure that the axis of microscope lens is ideally perpendicular to xy plane?
Hi Jeroen, I'm wondering if the mechanical issues with stitching multiple patterns could be overcome by some sort of feedback. Instead of slicing an image up into a certain number of pieces ahead of time, and trying to get the mechanics to move to a precise location, you could move the stage as precisely as possible, but then use the computer to adjust the image "on-the-fly" to account for any misalignment. A linear encoder not connected to the lead screw could provide you with a more precise location. If you can't get the mechanism to get you to a precise position, maybe you can use software to accurately project the right image for whatever position you're at? Of course this might cause additional problems. Off the top of my head, I'm not sure if double exposures would be an issue (if the images overlapped), or if the relatively low resolution of the projector would cause aliasing artifacts (if the misalignment is fractional compared to the size of a "pixel"). Anyway, this is incredible work, and I wish you great success.
Well, i have to say this is truly remarkable i'm self taught working in electronics, at some point when i have enough spare money i'd like to get into making projects like this really watching you and others do stuff like this is really really inspiring and not just makes me feel delighted and increase my curiosity, but also inspires me to aim higher on my goals hope i can see more of this in the future pd. also maybe try disabling automatic translation for the title and description if that on your side
I'm not sure which language you are using for subtitles. Only on the English I have influence, the rest is automatically generated. I know the translations suck, but in some cases they can help people to understand a certain issue a bit better.
@@HuygensOptics not the subtitles, the title of the video and the description, for some reason Google decided a few months ago it was a good idea to translate them and give the user no choice (at least I have found no way to revert that whatsoever and if I don't remember wrong I read on reddit that the only people who could deactivate that were the uploaders)
in view of the lead screw deformation problem, I think you could use a linear encoder for feedback control, instead of encoders on stepper-servo. the one i use in vision measuring system have a resolution of 0.5 micrometer.
Great video. Few suggestions how to solve things possibly cheaply. Add digital read-out on both axis. This way you can do a closed loop positioning. Use precomputed deviations for rough positioning and the DRO loop to position accurately. It will solve non-linearity. The squarness can be compensated using calibration (by using many known points on the good mask mounted in specific position on the stage) and proper control in software, by offsetting non-squarness out. Similarly the curves and other non-linear effects can be compensated out. The only thing that can't be fully compensated are rotations, which are of course also present in some of these non-linear deviations; and of course temperature dependent deviations (so keep things constant during calibration). I am not saying it is easy, or the easiest, but probably cheapest way to try. I see you are a capable programmer, so think about it. New xy motorized high accuracy / repetability (sub-micron) linear stage of medium size like this will still be expensive. About 9000$ from a quick research.
Thanks for the suggestions. So, squareness of the table can also be adjusted, I still need to do that, that is another 3 microns of error out the window. Yeah, if I had the money, I would probably just buy nano stages from Newport, PI, or Attocube. But then of course I would not learn as much as I do now. The best way to achieve very high accuracy would be to use an interferometric feedback I guess, using mirrors on the sides of the table. I'm looking into that, but I know on forhand that it will not be trivial to implement. Anyway, it's been a fun project so far.
Question / Comment @ 8:58 re non-linearity of the actuation. I would agree that imperfections in the lead screw would contribute. I wonder if you've considered the effects of microstepping... I know this is the case with steppers ( I recall you said servos -- unsure if still applicable ) You will get a cyclical ( which your graph seems to show ) deviation based on the electromagnetic power ... which is determined by the position relative to a full-step where the rotor is currently placed. Your seemingly cyclical deviation could be artifacts of the micro-stepping process -- especially when running super low currents.
Thanks for the suggestion but these imperfections are cyclic with a full rotation, not with a step. Also the servos have a 4000 steps/rev optical encoder and use a feedback loops to go to a specific rotational position.
Really cool! Have you looked into how distortions in the projected image affect the stitching? Perhaps you could measure the distortion field by projecting one of the MTF mapper test patterns onto the photoresist. Then you could maybe then compensate for the distortion in your software by applying the opposite distortion to the digital image to be projected. Also, because the objective lens is probably not optimized for UV, the projected image intensity may not be uniform across the field. Your software could compensate for that too. Not sure you can do anything about gradients in sharpness, but maybe that is less critical
Also, because most modern objectives are infinity corrected, I'm a bit surprised that you're not using a tube lens in the imaging system. Is this particular objective of the finite conjugate type? Otherwise, a tube lens could improve sharpness away from the center (by flattening the field), and should also help keep the illumination closer to telecentric at the image, which may help avoid sloped sidewalls
That is correct, best practice is using a tube lens with this objective. I added it later to the design, and also added a microscope cover slide correction and removed the cube beam splitter. Actually, I made quite a lot of changes to this initial setup, maybe I should discuss the optical design modifications in a future video.
Very interesting. Is it possible to add a feedback measurement system, like a laser distance meter onto the x and y axis to get rid of the leads crew deviations? Maybe using a trinamics stepper driver might reduce the servo noise.
you might get better results using a plain fluorite objective instead of the apochromatic objective- the apos have a bunch of additional optical elements to correct chromatic aberration, but you're only using narrow band UV so those extra elements are just going to reduce your resolution and you won't gain anything
Didn't one Ariadne rocket crash due to using imperial and metrical standards in the software? Btw. I really want a big telescope made with flat lenses, really cool.
Wow, you are really doing great work. Repeatable errors are your friend. Maybe you can make software corrections for all repaitable errors like imperfections in the screw and table alignment. Print a pattern over the full range of your system, have this measured very precice and compensate these errors with your software. You might be able to make your system more precise. (I studied precision engeneering in Eindhoven and did some work at ASML, now about hunders years ago. But I quickly discovered that I am not the person for um or nm. You sound like it is easy as bricklaying, I am impressed. Now I work in Zambia and it fills my hart with joy if our trainees reach precsion of +/- 1mm ;). Really impressive.
Well the solution I chose was to use optical encoders with 100nm precision. Feedback loops are in my opinion the best way to eliminate positional errors.
coreless linear motor + encoder at least 100nm resolution,checking axial squareness/sraightness/tureness via laser interferometer，generated related compensate table use motion control to compensate via table
really really enjoy your setup build up, ! i am currently using a 2P spectroscopy. i would like to build one myself, I require a Z stack, galvomirrors and xyz stepper motors for movement. which devices do you use for the Galvo mirrors?
you've addressed the drive train precision elegantly but the weak link seems to be the actual x-y stage. Is't the precision limited by the precision of the stage?
Why not just use a couple of laser diodes as interferometers - one for each x/y stage, and use the outputs from those as the 'step' signal with which to close the loop for each axis? OK, will need temperature stabilization for each laser diode (to keep the wavelength from shifting), but it should mean that the leadscrew nonlinearity doesn't matter...
@@HuygensOpticsinterferometer is optical and its pattern comes for free and you only need a single photodiode. And it works well for scanning. Too slow for steps
I was thinking about a less complex machine for exposing double layered PCB. It requires much less precision, but still needs to be very precise. One thing that puzzles me is how would one make it for double sided PCB, meaning exposing it from both sides. The back side placing must exactly match the position of the front. PCB shape can't be ideal. Also, PCB thickness is not the same (from one project to another). Maybe I'll come up with something after some more thinking and research. But does anyone seen something similar somewhere?
May you drill all vias first. Then take a photo with the CCD and bend your virtual wires before you illuminate the CCD. Most of your SMD stuff will be independent
Love your channel! For my understanding, the resolution of the axis sould be smaller then the projected pixel size? Because how could you ensure that the pixel sized features are stiched correctly together?
That is correct. However, it is not something that I have achieved yet. I am in the process of modifying my hardware, but it is doubtful that even in the new configuration I will reach < 0.3um position accuracy. I guess I would have to get my hands on a real wafer table from a wafer stepper to achieve this.
@@HuygensOptics In my expirience with building position stages, the problem lies in the mechanics itself and the lack of precision position meausurement. I don't have a interferometer :) Even with that and a closed looped driver and the best motors I couldn't archive sub mircon levels. Repeatability and error where in the +/- 1µm range. Nice thermo drifts over the day where common, in the end they were the easiest think to correct for. I was just giving up as I found a solution. The buzz word is flexures. ruclips.net/video/PaypcVFPs48/видео.html The best think you can even print this thinks: gitlab.com/openflexure/openflexure-block-stage Diy flexure piezzo stage: www.dpreview.com/forums/thread/4441997
Thanks for the links Jonas. The flexture stage shown in your last link is exactly the one I used for my project. So one of the issues with flextures is their very limited range. So you need a combination of normal steppers and piezo flextures to make multi mm size patterns, which makes makes control complex. If I had sufficient funds for that, I would just use piezo-electric drives: www.physikinstrumente.nl/en/technology/piezoelectric-drives/
Really enjoyed these two videos. Interesting how you can use this method to focus light with a flat element or lens. Nice results. Would these lens focus a laser beam? Thanks!
Really cool project! probably a question for the last video but how do you collimate the output of the LED? I thought that was incredibly hard as it emits over quite a large area
Actually, I think it is an advantage that the beam is not perfectly collimated, because even with an LED I observe intensity variations in the beam due to diffraction effects on the DMD-chip. So I roughly colimate the LED-light using a lens and then I use a small concave mirror to direct the light into the microscope objective via the DMD-chip. Its a bit like I describe in the video linked below around 8 minutes, but then in reflection in stead of transmission: ruclips.net/video/E9uU2h1Ldcg/видео.html . Hope this clarifies it.
@@HuygensOptics I have seen that video actually, also really interesting, just didn't realise how hard it is to get even illumination. I guess you get problems from the shadow of the bond wire on the led chip?. But maybe I'm missing some fundamental optics here, I thought the beam had to be as close to collimated as possible to image the 'mask' onto the focal plane of the lens?
I think the linearity error comes from the positioning accuracy of your servo motors. Servo error should not be cumulative and I think it shows on your graphs that the error follows the steps, sine/cosine sort of wave form. Plus the lead screw error...
Random question (from a non-expert), could the periodic oscillating error of be down to a misalignment of the servo with the micrometer shaft (i.e. something solved with a constant velocity coupling)? Anyways, amazing content and even more amazing that it doesn't generate more views. Thanks for sharing.
That is an excellent idea. Part of the non-linear behaviour of the lead screw could be caused by the external forces on it from (for example) the motor axis. Maybe a constant velocity coupling can help tom minimize these forces. Thanks!
Damn! Your videos are so very interesting that I could not tear myself from RUclips until 5:oo in the morning! You need to include a warning about how addictive your videos are!
maybe I should rename the channel to "heroin optics" then?
Loll
@@HuygensOptics do so hahaha
Great stuff, really impressive machine! I agree with the leadscrew analysis (essentially the same periodic wormgear behavior that telescopes calibrate out), but wanted to throw out another potential contributor: motor cogging. E.g. the stator slots prefer to settle into low-energy positions relative to the permanent magnets and experience a force when moving in/out of those positions. Nicer servos will use tricks like skewing the magnets to reduce cogging (or ironless motors which have no cogging, but also very poor torque). It can also be compensated in software by building a "cogging map" which commands the motor to use more or less current at specific locations in the revolution so that cogging is nullified. The periodicity depends on the number of slots/magnets but will be higher than just once-per-rev, so is probably a higher-frequency signal overlaid the 4000-step non-linearity chart you showed.
You deserve a pat on the back, even real steppers suffer from many on the same issues and have a variety of software calibration files to account for this.
Very impressive! I admire how far you got in such a short time!
When I worked for ASML in 1998-99 as a software engineer, I learned that there are some very weird problems that most people might not think about. The machines i worked on had 23 degrees of freedom, moving a 30kg marble slab with a wafer around at up to 6m/s or so. Of course the reticle had to move in opposite direction (but at 1/4th the speed IIRC) and there were counter-weights to keep the machine in place.
The motion control of the linear actuators needed to compensate for "cogging" (i.e. variation in the magnetic field depending on where the magnets are), and things like the bending and un-bending of the electric cables that are attached to the stage. They used a Hewlett-Packard laser interferometer to measure the position of the wafer stage but I think that was just for rough estimates: the wave length of a red laser is much longer than the accuracy of the machine (180nm in that time).
The software was responsible for basically keeping the speed constant, so it gives the motor a big push, then lets the stages "coast" to the other side with just a little bit of compensation for irregularities along the way. Then at the end of the track the motor would be pushed in opposite direction to stop the mass from moving.
And of course it's "fun" if something fails and the pushback never happens; when a wafer stage crashes, it will cause significant damage that will cost a lot of time and money to repair. I remember someone telling me that in a fab, even opening a side panel causes so much disturbance to the air flow that it took an entire day for things to settle and make the machine usable and productive again.
What they achieve at ASML (especially the new machines) is absolutely incredible.
Former ASML software engineer here, 2002-2009/
Yeah, we've created some pretty crazy stuff there, proud to have been part of the team
@@barmalini I just started as a development engineer at ASML's German optics supplier and am continually amazed at the engineering they do (although our stuff isn't too bad either).
I worked in Texas Instruments Optics Department in Dallas Tx, where we perfected the coating for the top lens of the digital light chip. After our Optics department made the coating repeatable, TI moved the coating to SE Asia.
Ah Yes, outsourcing the tech we need for national security (physically, and financially)
🤣🤣😅🥲😢😭😭😭😭😭🤬🤬🤬
@@gaussdog ah, yes... national security... having military presence all over the globe... 🤦♂️ more guns means more security, right?
I worked Dmat on DMD South in the building for the first 12 years at T.I.
I did not know that the lens is integrated with the chip.
6:25 - I want to mention that "Trinamics" driver chips produces no noise and is pure analog signal to drive the servo/stepper motors, really good for this micro stepping to remove any unwanted vibrations.
Oh no, Trinamic stepper driver ICs still make lots of noise, but humans can't hear their ultrasonic PWM frequency. Yes, they're great chips, and yes their 35kHz (configurable) hard-switched PWM is adequate for most purposes, but sadly no they're definitely not pure analog and definitely not vibration-free in general. If you can find, design, or hack a PCB to use an 8MHz external clock for the stepper driver IC, then you can configure it to an audible 15.6kHz chopper frequency and listen with your own ears.
I definitely don't work for Trinamic and I only have direct experience designing around the TMC2209 and TMC5160 ICs. If you've found a Trinamic IC that is truly analog, or can generate >100kHz PWM to allow efficient external filtering (without overclocking outside maximum ratings), I'd love to hear about it.
@@W77W well I would use trinamics for this project since it eliminates noises compared to the drivers he is using at the moment. But I’m not sure if ultrasonic noise makes no difference compared to the lower frequency noise which humans can hear. Trinamics are not pure analog but they are very close to it. I can’t think of any better driver chips other than trinamics. Another problem is that I would not use stepper motor, they are no good for 1 micron precision.
Servo motors with close loop is better for precision positioning and I would also include an optical encoder with incremental accuracy of 1 micron.
Trinamics TMC4671 can switch at 100khz. Here is the link:
www.trinamic.com/company/news/news-detail/implement-your-servo-controller-in-a-day-with-the-tmc4671/
@@ShopperPlug Thanks for the quick reply and thanks for the link, I agree with you 100% that the Trinamics are the best single-chip solutions available (to my knowledge). I just wanted to throw in my 2 cents of caution because I did have an issue with a PCB drilling machine, where the 0.1mm micro-drill bits were breaking because the XY stage was vibrating ultrasonically enough to flex the drill bit, even when all axes and spindle were "stationary".
@@W77W Wow 😮, that is really interesting, how did you come up with the determination of the fact that the Trinamic’s ultrasonic switching is what caused the 0.1mm drill bits to flex and break?
Would like to see your project if you documented it, I have plans to build a precision drill machine for PCB as well.
Right now I’m gathering plans and materials for building a precision sub-micron laser PCB photoresist sensitizer.
I’m planing to use custom built linear induction motor, no servo or stepper motor.
I will use heavy triple-A above laboratory grade precision granite for the base.
I’m also considering to use air bearings.
This is the basic design I’m planning to use for the CNC laser sensitizing system, it is used by all high precision companies that provides nanometer positioning accuracies, and ultrasonic vibrations does not affect the x/y system since it is 100% frictionless (not considering the air molecules) when driving the linear induction motors:
static.pi-usa.us/fileadmin/_processed_/7/d/csm_Air-Bearing-Planar-Stage_A-322_400w_ed5e39796f.jpg
@@ShopperPlug I'm really sorry but it's not my machine, so I don't get to share the build.
I can say that epoxy granite (en.wikipedia.org/wiki/Epoxy_granite www.adambender.info/post/2017/03/25/epoxy-granite-machine-frame-how-to) is amazing, superior to the natural material in many respects. Being able to pre-cast notches, rails, and screw threads is really nice. Of course you need something flat to start with, but you can cast your epoxy granite upside-down on your surface plate to transfer almost all the flatness properties from one to the other. In our case we first stuck a precision parallel slightly above the surface plate, then made the casting to generate a precisely parallel-sided rectangular slot in the base, connected to the surface by an imprecise window, so it's roughly an inverted T-shaped cross-section like a milling machine table's T slots. I don't know if the thermal expansion properties are good enough for nanometer precision, but they're certainly good enough for a drilling machine. Just don't cast your very expensive parallel into a block of epoxy granite without a solid plan to get it back out. In our case we just chilled it to shrink the steel.
It's also surprisingly easy to make DIY air bearings (ruclips.net/video/K_N_h_mKf-4/видео.html). Machining graphite is trivial, so you can easily make exotic bearing configurations. For example, you could machine a roughly X-shaped piece of graphite very slightly oversize for the parallel-sided T-slot, squeeze the X so it flexes very slightly, and wedge it into the slot (rather like installing pistons in an engine). Now you have an air bearing axis that's as precise as the parallel you used to cast the slot, and it's fully dimensionally constrained by the 4 sides of the slot, at least up to the elastic stiffness of graphite. You just mount the next axis directly to the graphite block. I suppose if you need a really large travel you might want two parallel slots so you don't have to buy/make a huge precision parallel or a huge chunk of graphite, but the idea is the same. Again I'm not certain you'll get nanometer precision out of this, but for a drilling machine it's great.
If you can't follow my description I can draw up a sketch.
I connected the dots to the ultrasonic resonance by several rounds of troubleshooting. I started by observing that on smaller PCBs (~5*10mm) the drill bit would snap very regularly, often at the same location on the PCB. Larger PCBs were immune. I figured that something must be moving, but I had no idea if the spindle had weirdly inconsistent runout, or the stage wasn't stiff enough, or the PCB was flexing, or maybe thermal expansion or air/vacuum pressure variations. I made several wrong guesses, then eventually hooked up a cheap continuity probe from the tool to the PCB to see if there was low-speed drift (thermal expansion driven). I use this probe all the time on my milling machine, so I was really confused when instead of a sharp on/off at the edge of the PCB, the LED would slowly get brighter as I bumped into the edge. Again some more wrong guesses, but I eventually hooked up an oscilloscope to the continuity probe, and clear-as-day you could see the 23kHz vibration on the 'scope. There's no way 23kHz was coming from outside the machine, so I pretty quickly followed it back to the source. Dialing up the PWM frequency and decreasing the holding torque solved the problem, but a heavier stage would also have helped.
You can turn the lense from the beamer backwards to demagnify the image. That is what sam Zoeloff did for his mask projector.
That could certainly work. A pity I did not think of that!
I tried that and doesn't work very well. After that you also need to pass the image through the microscope objective. You add more and more lens. I got the best results in my test with Huygens metod 👌
Probably the best channel for information. Keep it coming, you are doing great work.
Not sure how one of your lens grinding appeared in my feed but what you are dong is fascinating i'm so intrigued by the level of detail you have gone to in each stage of this project is truly masterful. I was considered building a homemade telescope and was stunned by the amount of work that goes into hand grinding a lens. Your solutions are well researched and executed and captivating to watch. Thank you for shearing.
Thank you RUclips for showing me your fascinating channel. I particularly enjoy watching people build equipment that would cost thousands and thousands out of various bits and pieces. You are doing this in a field that is both rare on RUclips and of particular interest to me. Thanks!
Wow. I have mechanical noise from much larger leadscrews when 3D printing- and I only am dealing with movements around 1/10th of a millimeter.
This kind of stuff seems orders of magnitude more difficult.
This is my favorite project so far. Amazing. Keep up the good work.
I used to work on the product team that made exactly what you need to get this working! We made twin axis heterodyne interferometers for these applications. Don't seem to be many of them floating around second hand though!
Hi Jeroen, I'm so glad that I found your channel. Also, I just wanted to let you know that diffraction is a classical phenomenon, not a quantum mechanical one (3:19). Thank you for making these videos!
Thanks. Btw: diffraction is also quantum if you consider it the result of a wave function instead of a classical wave.
@@HuygensOptics Thank you for your reply. It seems though having merely the wave property does not make it a quantum phenomenon. In fact, nothing in the theory of diffraction states that light is quantized, which is another necessary property for a quantum phenomenon. Maybe if you had a single photon emitter and you observed interference then you can say it's a quantum phenomenon. If you look at it as a Venn diagram, geometrical optics is a subset of wave optics (diffraction), which is a subset of quantum optics. In reality, quantum optical phenomena such as optical tweezers and Rabi Oscillations are much more difficult to observe. I'm sure you understand the matter very well, however I just don't want the viewers to get confused.
Very interesting. Thank you! You did not talk about influence of temperature on the device dimensions. Do you consider it as a problem? Aluminium expands ca. 0.1 percent over 50 oC.
How do you make sure that the axis of microscope lens is ideally perpendicular to xy plane?
Hi Jeroen, I'm wondering if the mechanical issues with stitching multiple patterns could be overcome by some sort of feedback. Instead of slicing an image up into a certain number of pieces ahead of time, and trying to get the mechanics to move to a precise location, you could move the stage as precisely as possible, but then use the computer to adjust the image "on-the-fly" to account for any misalignment. A linear encoder not connected to the lead screw could provide you with a more precise location.
If you can't get the mechanism to get you to a precise position, maybe you can use software to accurately project the right image for whatever position you're at?
Of course this might cause additional problems. Off the top of my head, I'm not sure if double exposures would be an issue (if the images overlapped), or if the relatively low resolution of the projector would cause aliasing artifacts (if the misalignment is fractional compared to the size of a "pixel").
Anyway, this is incredible work, and I wish you great success.
The best way indeed is to use accurate positional feedback to overcome mechanical flaws of the stage.
Congrats on getting it actually to work! This lenses look already nice. :)
Well, i have to say this is truly remarkable
i'm self taught working in electronics, at some point when i have enough spare money i'd like to get into making projects like this
really watching you and others do stuff like this is really really inspiring and not just makes me feel delighted and increase my curiosity, but also inspires me to aim higher on my goals
hope i can see more of this in the future
pd. also maybe try disabling automatic translation for the title and description if that on your side
I'm not sure which language you are using for subtitles. Only on the English I have influence, the rest is automatically generated. I know the translations suck, but in some cases they can help people to understand a certain issue a bit better.
@@HuygensOptics not the subtitles, the title of the video and the description, for some reason Google decided a few months ago it was a good idea to translate them and give the user no choice (at least I have found no way to revert that whatsoever and if I don't remember wrong I read on reddit that the only people who could deactivate that were the uploaders)
@@HuygensOptics Also off topic: any book you recommend to get started in optics?
@02:00
Don't beat yourself up over the contamination or anything!!!
One step at a time!
in view of the lead screw deformation problem, I think you could use a linear encoder for feedback control, instead of encoders on stepper-servo. the one i use in vision measuring system have a resolution of 0.5 micrometer.
Great video. Few suggestions how to solve things possibly cheaply. Add digital read-out on both axis. This way you can do a closed loop positioning. Use precomputed deviations for rough positioning and the DRO loop to position accurately. It will solve non-linearity. The squarness can be compensated using calibration (by using many known points on the good mask mounted in specific position on the stage) and proper control in software, by offsetting non-squarness out. Similarly the curves and other non-linear effects can be compensated out. The only thing that can't be fully compensated are rotations, which are of course also present in some of these non-linear deviations; and of course temperature dependent deviations (so keep things constant during calibration). I am not saying it is easy, or the easiest, but probably cheapest way to try. I see you are a capable programmer, so think about it. New xy motorized high accuracy / repetability (sub-micron) linear stage of medium size like this will still be expensive. About 9000$ from a quick research.
Thanks for the suggestions. So, squareness of the table can also be adjusted, I still need to do that, that is another 3 microns of error out the window. Yeah, if I had the money, I would probably just buy nano stages from Newport, PI, or Attocube. But then of course I would not learn as much as I do now.
The best way to achieve very high accuracy would be to use an interferometric feedback I guess, using mirrors on the sides of the table. I'm looking into that, but I know on forhand that it will not be trivial to implement. Anyway, it's been a fun project so far.
@@HuygensOptics small steps. It is already looking very promising and usable for some tests. ;)
Question / Comment @ 8:58 re non-linearity of the actuation.
I would agree that imperfections in the lead screw would contribute.
I wonder if you've considered the effects of microstepping...
I know this is the case with steppers ( I recall you said servos -- unsure if still applicable )
You will get a cyclical ( which your graph seems to show ) deviation based on the electromagnetic power ... which is determined by the position relative to a full-step where the rotor is currently placed.
Your seemingly cyclical deviation could be artifacts of the micro-stepping process -- especially when running super low currents.
Thanks for the suggestion but these imperfections are cyclic with a full rotation, not with a step. Also the servos have a 4000 steps/rev optical encoder and use a feedback loops to go to a specific rotational position.
Really cool! Have you looked into how distortions in the projected image affect the stitching? Perhaps you could measure the distortion field by projecting one of the MTF mapper test patterns onto the photoresist. Then you could maybe then compensate for the distortion in your software by applying the opposite distortion to the digital image to be projected. Also, because the objective lens is probably not optimized for UV, the projected image intensity may not be uniform across the field. Your software could compensate for that too. Not sure you can do anything about gradients in sharpness, but maybe that is less critical
Also, because most modern objectives are infinity corrected, I'm a bit surprised that you're not using a tube lens in the imaging system. Is this particular objective of the finite conjugate type? Otherwise, a tube lens could improve sharpness away from the center (by flattening the field), and should also help keep the illumination closer to telecentric at the image, which may help avoid sloped sidewalls
That is correct, best practice is using a tube lens with this objective. I added it later to the design, and also added a microscope cover slide correction and removed the cube beam splitter. Actually, I made quite a lot of changes to this initial setup, maybe I should discuss the optical design modifications in a future video.
Very interesting. Is it possible to add a feedback measurement system, like a laser distance meter onto the x and y axis to get rid of the leads crew deviations?
Maybe using a trinamics stepper driver might reduce the servo noise.
you might get better results using a plain fluorite objective instead of the apochromatic objective- the apos have a bunch of additional optical elements to correct chromatic aberration, but you're only using narrow band UV so those extra elements are just going to reduce your resolution and you won't gain anything
Very cool! Can't wait for more videos!
Didn't one Ariadne rocket crash due to using imperial and metrical standards in the software? Btw. I really want a big telescope made with flat lenses, really cool.
How did you measure the displacement so accurately?
Really cool project! Great video!
Wow, you are really doing great work. Repeatable errors are your friend. Maybe you can make software corrections for all repaitable errors like imperfections in the screw and table alignment. Print a pattern over the full range of your system, have this measured very precice and compensate these errors with your software. You might be able to make your system more precise. (I studied precision engeneering in Eindhoven and did some work at ASML, now about hunders years ago. But I quickly discovered that I am not the person for um or nm. You sound like it is easy as bricklaying, I am impressed. Now I work in Zambia and it fills my hart with joy if our trainees reach precsion of +/- 1mm ;). Really impressive.
Well the solution I chose was to use optical encoders with 100nm precision. Feedback loops are in my opinion the best way to eliminate positional errors.
@@HuygensOptics even better.
coreless linear motor + encoder at least 100nm resolution,checking axial squareness/sraightness/tureness via laser interferometer，generated related compensate table use motion control
to compensate via table
Great video! What is the etcher that did i used?
Is it possible to make a tinier focus lens for the stepper with the stepper? Guess not.
Hmm i’d prefer ball screws..
really really enjoy your setup build up, ! i am currently using a 2P spectroscopy. i would like to build one myself, I require a Z stack, galvomirrors and xyz stepper motors for movement. which devices do you use for the Galvo mirrors?
you've addressed the drive train precision elegantly but the weak link seems to be the actual x-y stage. Is't the precision limited by the precision of the stage?
Correct, the solution is to add high-accuracy encoders and use a feedback loop.
Why not just use a couple of laser diodes as interferometers - one for each x/y stage, and use the outputs from those as the 'step' signal with which to close the loop for each axis? OK, will need temperature stabilization for each laser diode (to keep the wavelength from shifting), but it should mean that the leadscrew nonlinearity doesn't matter...
Couldn't you improve the linearity of the lead screws using feedback with an interferometer?
Yes, that is the way to do it, use position feedback. Btw, optical encoders are a cheaper and easier way to do it then interferometers.
@@HuygensOpticsinterferometer is optical and its pattern comes for free and you only need a single photodiode. And it works well for scanning. Too slow for steps
I was thinking about a less complex machine for exposing double layered PCB. It requires much less precision, but still needs to be very precise. One thing that puzzles me is how would one make it for double sided PCB, meaning exposing it from both sides. The back side placing must exactly match the position of the front. PCB shape can't be ideal. Also, PCB thickness is not the same (from one project to another). Maybe I'll come up with something after some more thinking and research. But does anyone seen something similar somewhere?
May you drill all vias first. Then take a photo with the CCD and bend your virtual wires before you illuminate the CCD. Most of your SMD stuff will be independent
"Before I get too excited and have a little accident." -My favorite line
Love your channel! For my understanding, the resolution of the axis sould be smaller then the projected pixel size? Because how could you ensure that the pixel sized features are stiched correctly together?
That is correct. However, it is not something that I have achieved yet. I am in the process of modifying my hardware, but it is doubtful that even in the new configuration I will reach < 0.3um position accuracy. I guess I would have to get my hands on a real wafer table from a wafer stepper to achieve this.
@@HuygensOptics In my expirience with building position stages, the problem lies in the mechanics itself and the lack of precision position meausurement. I don't have a interferometer :) Even with that and a closed looped driver and the best motors I couldn't archive sub mircon levels. Repeatability and error where in the +/- 1µm
range. Nice thermo drifts over the day where common, in the end they were the easiest think to correct for. I was just giving up as I found a solution. The buzz word is flexures. ruclips.net/video/PaypcVFPs48/видео.html
The best think you can even print this thinks:
gitlab.com/openflexure/openflexure-block-stage
Diy flexure piezzo stage:
www.dpreview.com/forums/thread/4441997
Thanks for the links Jonas. The flexture stage shown in your last link is exactly the one I used for my project. So one of the issues with flextures is their very limited range. So you need a combination of normal steppers and piezo flextures to make multi mm size patterns, which makes makes control complex. If I had sufficient funds for that, I would just use piezo-electric drives: www.physikinstrumente.nl/en/technology/piezoelectric-drives/
Really enjoyed these two videos. Interesting how you can use this method to focus light with a flat element or lens. Nice results. Would these lens focus a laser beam? Thanks!
Yes they do. See: ruclips.net/video/uf3Y0-6NbjQ/видео.html at around 3 minute 30.
seems like a job for flexure mechanisms and linear motors
Really cool project! probably a question for the last video but how do you collimate the output of the LED? I thought that was incredibly hard as it emits over quite a large area
Actually, I think it is an advantage that the beam is not perfectly collimated, because even with an LED I observe intensity variations in the beam due to diffraction effects on the DMD-chip. So I roughly colimate the LED-light using a lens and then I use a small concave mirror to direct the light into the microscope objective via the DMD-chip. Its a bit like I describe in the video linked below around 8 minutes, but then in reflection in stead of transmission: ruclips.net/video/E9uU2h1Ldcg/видео.html . Hope this clarifies it.
@@HuygensOptics I have seen that video actually, also really interesting, just didn't realise how hard it is to get even illumination. I guess you get problems from the shadow of the bond wire on the led chip?. But maybe I'm missing some fundamental optics here, I thought the beam had to be as close to collimated as possible to image the 'mask' onto the focal plane of the lens?
I'm curious, how did you measure the step size of 11nm for the piezo actuator?
By measuring the full dynamic range and divide by 4095 steps (12-bit resolution)
Didn't see that Spinal Tap quote coming! LMAO😂
By the end of the world, you are the guy that I want to stay with….
I think the linearity error comes from the positioning accuracy of your servo motors. Servo error should not be cumulative and I think it shows on your graphs that the error follows the steps, sine/cosine sort of wave form. Plus the lead screw error...
Random question (from a non-expert), could the periodic oscillating error of be down to a misalignment of the servo with the micrometer shaft (i.e. something solved with a constant velocity coupling)?
Anyways, amazing content and even more amazing that it doesn't generate more views.
Thanks for sharing.
That is an excellent idea. Part of the non-linear behaviour of the lead screw could be caused by the external forces on it from (for example) the motor axis. Maybe a constant velocity coupling can help tom minimize these forces. Thanks!
super cool!
You need a over pressurized chamber or compartment to drive out the dust.
What resolution it will be?
the word ‘Maskless’ didn’t age well xD
great videos, thank you for sharing your knowledge and experience
Following in Sam Zeloof's footsteps I see
can you granite ?
Maybe
Unfortunately I can only give 1 like…
haha, you need a clean room for better contamination control
I want to be you when I grow up.
Who add Chinese subtitles? I guess its CCP did