<glowplug>
I'm not the only one in here! Thats fantastic. =)
<glowplug>
Is this project still active?
<azonenberg>
It's slow, i've been the main one working on it and i've been busy with my "real" research
<azonenberg>
The evaporator on campus that i had been using was down for repairs for a while but is up again
<azonenberg>
then i need to find time to upgrade my spin coater
<glowplug>
I must say that you are amazing sir. This is very important research and work and I want to thank you personally for what you have accomplished and shared. 8)
<glowplug>
Are you familiar with the OpenCores / OpenRisc projects?
<azonenberg>
I am
<azonenberg>
Though homebrew fab is a LONG way from being able to do that
<azonenberg>
the current state of the art is Jeri's millimeter-sized CMOS inverter and my 20-micron lithography without any working transistors (thoguh i haven't tried, been focusing on the litho)
<glowplug>
I was very impressed by the 20-micron lines. I couldn't believe it.
<azonenberg>
Lol
<azonenberg>
I think i can get a lot better than that with some time and effort
<glowplug>
Without this technology progressing OpenCores will stagnate on proprietary FPGA chips running at 50mhz. Not a good place to be.
<azonenberg>
A paper from SDSU demonstrates a UV direct write laser lithography system for under $1K
<azonenberg>
with better than 5um resolution across a large area
<glowplug>
I was just going to ask if you had tried direct laser yet. 8)
<azonenberg>
basically a blu-ray diode with a blanking circuit, a preciison x-y stage, and some control hardware/software
<azonenberg>
It's very high on my list of things to build
<azonenberg>
But i have other stuff going on so it hasn't happened yet
<glowplug>
A cartesian blu-ray machine. Thats trivial. How much better than 20 microns do you think such a setup can achieve?
<azonenberg>
That paper quoted 2um "easily"
<azonenberg>
I was thinking you could then do a second shrink step
<glowplug>
Thats almost an Intel 386.
<azonenberg>
Make a 2um mask and then use it in a reducing stepper
<azonenberg>
with say another 4x reduction
<azonenberg>
so direct write for photomask making at 4x your desired feature size
<azonenberg>
then projection litho through the mask
<glowplug>
Do you have a link to the PDF?
<azonenberg>
Sadly no
<azonenberg>
i printed it out and the title was messed up
<azonenberg>
it's by Kassegne and Khosla
<azonenberg>
automated maskless photolithography using UV laser diodes
<azonenberg>
should be able to google it
<glowplug>
I will try to track it down. I have a CNC, its homebuilt but decently accurate. I don't think it would complain about having a blu-ray diode strapped on.
<azonenberg>
It looks like the limitation in their system is twofold
<azonenberg>
first, x-y positioning
<azonenberg>
second, spot size
<glowplug>
I am extremely surprised that there aren't more contributing. I'm sure many more people understand how incredibly important this is.
<azonenberg>
mechanical and optical respectively
<azonenberg>
I would like to get most of the equipment scaled to glove-box size
<azonenberg>
to avoid needing a full cleanroom
<glowplug>
Both are difficult problems to solve.
<azonenberg>
or alternatively, sealed cassettes of some sort to move wafers between machines
<azonenberg>
The nice thing about larger features is that you're less sensitive to particles
<glowplug>
An accurate cartesian robot in a glovebox is non-trivial. Really servos should be used not steppers, accurate encoders are big, servos are big.
<azonenberg>
One idea i had for positioning was interferometry
<glowplug>
I think that sealed cassettes is more realistic.
<glowplug>
But still that doesn't answer why there aren't more people helping you. Haha
<azonenberg>
Lol
<azonenberg>
You forget, the majority of hobbyists are scared of soldering BGAs
<azonenberg>
or even writing HDL
<azonenberg>
if you tried to show them CMOS layout they'd run away and scream
<glowplug>
I'm trying to learn Verilog right now to contribute to the mor1kx core that Stefan and Julius are working on.
<soul-d>
or able to fix a fancy lab in their backyard
<azonenberg>
whereas i'm the kind of guy who has nude photos of MCUs on his bedroom wall (PIC12F683 metal 3 in particular)
<glowplug>
Haha!
<azonenberg>
glowplug: in case you haven't checked out siliconpr0n.org
<azonenberg>
do so
<azonenberg>
i'm one of the two main people behind the group
<azonenberg>
i think this is M5 and M6 but the polishing was a little uneven so i can't be sure
<azonenberg>
I do know it's the smallest pitch on the device by comparing it to cross sections from chipworks, it's a lot smaller than the top two or three layers
<glowplug>
So let me get this strait. If we have a very accurate cartesian robot wth a blu-ray diode we can build 2um semiconductor devices using techniques that you already know?
<azonenberg>
glowplug: Jeri Ellsworth has a FEOL process for doing transistors working at macro-scale without lithography
<azonenberg>
If you can do direct-write exposure, you can buy standard photoresist and developer solutions and do lithography
<azonenberg>
The last remaining problem is interconnect
<soul-d>
what am i looking at azonenberg looks like egyptians beat you to nano scale to
<azonenberg>
I have no idea what this circuitry does, i did no analysis - the point was to see if the SEM had the resolution for a modern deep submicron process
<glowplug>
How much work needs to be done from a design perspective to get functional logic gates?
<azonenberg>
Apparently it does, this is Samsung 45nm
<azonenberg>
glowplug: Alliance etc have standard cell libraries if we can figure out how to build them
<azonenberg>
Jeri has the transistor angle mostly figured out
<azonenberg>
I've made good progress on litho but it's not done
<azonenberg>
The last remaining problem is interconnect
<glowplug>
Should we ignore the direct laser methods due to their limitations in process accuracy?
<azonenberg>
Jeri's stuff used conductive epoxy and regular fine-gauge hookup wire
<azonenberg>
No
<azonenberg>
Definitely not
<azonenberg>
I think that's the way to go
<glowplug>
The theoretical smallest device is at the wavelength of the laser itself correct?
<azonenberg>
For direct write, approximately yes
<glowplug>
Which is determined by the photoresist we can acquire.
<azonenberg>
Well, it's going to be probably mercury vapor H-line (405nm)
<glowplug>
~250nm might be possible with direct laser assuming photoresist at those wavelengths is avialable.
<azonenberg>
which is what most UV diodes use
<glowplug>
405nm. Thats not bad at all either.
<azonenberg>
Deep UV photoresists are not sold to the general public
<azonenberg>
193nm is the standard wavelength for most mass production afaik
<azonenberg>
You can buy them if you're Intel
<azonenberg>
By the 55 gallon drum
<soul-d>
k cool ,( forgot i was cooking for a bit :P )
<glowplug>
Exactly. Thats the main problem with direct laser.
<glowplug>
Availability of the resist.
<azonenberg>
but regular resists are relatively easy to get hold of
<glowplug>
WIth 405nm we are looking at theoretical devices with frequencies 10x higher than FPGA's. I think OpenCores expirimentors would be quite happy with that.
<azonenberg>
Shipley 1813 can be bought by the quart from fisher scientific for $372 for example
<azonenberg>
If a bunch of us split a bottle that's quite reasonable
<glowplug>
You have really done your research. =)
<azonenberg>
Then you can get HMDS from SPI supplies for $12 for 30ml
<azonenberg>
or $15 for 100ml but the larger volume needs hazmat shipping
<azonenberg>
Which is why they offer the small option
<azonenberg>
MTI sells single wafers
<azonenberg>
including with oxide/nitride films
<azonenberg>
(though you're limitd by what they have in stock)
<soul-d>
chemical sharing you will hit stuff like shipping :P probably need to live close/same country
<azonenberg>
Most other wafer sellers wont touch you if you buy less than a boat of 25
<azonenberg>
Same country would be nice
<glowplug>
So what is immediately necessary to progress is a CNC with ~405nm stepping accuracy.
<azonenberg>
Not even that
<azonenberg>
1um is plenty for a lot of stuff
<azonenberg>
I want to put the initial focus on MEMS
<azonenberg>
Get the lithography down
<azonenberg>
But forget about trace metal contamination
<azonenberg>
You can use much less pure chemicals for MEMS
<azonenberg>
you care about particles but not ions
<glowplug>
True. For research and progress it's not necesary to be at the theoretical smallest. We could have football field sized devices as long as we are figuring these things out.
<azonenberg>
injectorall.com sells photoresist in small volumes intended for PCBs
<glowplug>
My CNC has 1um accuracy as do most.
<azonenberg>
It's probably not CMOS grade
<azonenberg>
But for MEMS it'll do just fine
<azonenberg>
That's what i did all of my testing with
<glowplug>
Have you had any interest from hackerspaces or other organizations like that to help with the research?
<azonenberg>
I haven't had the time to push too hard on recruitment
<azonenberg>
Once I graduate I'll "only" be working 40-50 hour weeks
<glowplug>
It's disturbing to me that people aren't throwing equipment and money at you.
<azonenberg>
and thus have weekends etc to play on stuff
<azonenberg>
But i've been funding my operations out of a grad student's stipend
<glowplug>
If you don't mind me asking where are you located? I'm in Michigan.
<azonenberg>
RPI
<azonenberg>
Working on my PhD (in computer science, go figure)
<glowplug>
Ahh New York.
<azonenberg>
Then some of the folks here are in europe i think
<azonenberg>
wolfspraul, whenever he shows up, is in china
<glowplug>
Does anyone else except you have physical equipment?
<azonenberg>
oh, and B0101 is in Singapore
<azonenberg>
She's done some minimal lab work
<azonenberg>
But not much
<azonenberg>
Most of the others just talk
<azonenberg>
Sync__: is a grad student somewhere and has equipment but idk how much he's done toward home fab
<azonenberg>
In my case I have a home lab plus i'm on good terms with the manager of the electron microscopy lab on campus
<azonenberg>
I did all of my metal coating in their evaporator
<azonenberg>
since i don't have vacuum gear at home yet
<azonenberg>
So like the nyan cat on the website
<glowplug>
How much is the evaporator?
<azonenberg>
that was evaporated copper using their system
<soul-d>
last time i posted a pic of my "lab" azonenberg had to cry :S
<glowplug>
Haha. Yeah my "lab" is a color laser printer and DIY cnc. :/
<soul-d>
cnc can build things
<azonenberg>
glowplug: Upper bound? A commercial small-sample sputter coater
<azonenberg>
If you homebrewed, you could likely make it a LOT cheaper
<glowplug>
Homebrew is more what I was thinking.
<glowplug>
If its $13k retail then I'm going to assume $500 for DIY.
<azonenberg>
I'd budget more like $1K but we'll see
<azonenberg>
If you want to get serious about this kind of stuff, i'll set you up with wiki access
<azonenberg>
PM me your google code email address
<azonenberg>
and i'll add you as an editor
<azonenberg>
you can start making pages with hardware designs
<azonenberg>
we also need wiki pages with collections of relevant papers
<glowplug>
I try to stay away from google technology whenever possible. But documentation is something that I can contribute.
<glowplug>
Have you considered a github page?
<azonenberg>
I was planning to make a dedicated website for it at some point with a mediawiki
<azonenberg>
If you bug me in a week or so i might get to it
<glowplug>
How about this. Let me toss up a site on gitpages then email you a link then tell me what you think about it. If its not garbage I can fork it to a shared repo (not my personal one) that way more people can add content ect.
<azonenberg>
What about mediawiki?
<azonenberg>
do you have something against that?
<azonenberg>
I just want to move to my own domain at some point rather than bouncing between third-party services and having content scattered around
<glowplug>
Unfortunately it has an SQL backend. I can design a quick HTML5 page that can survive without any special database and host for free on github.
<azonenberg>
I have several VPSes
<glowplug>
If you have a server with SQL and everything setup that would work too I suppose. =)
<azonenberg>
and i own a couple of domains
<azonenberg>
it'd be the work of an hour or two on the weekend to set up a dedicated domain for it and put up a wiki
<azonenberg>
That way i can move the stuff around and as long as i own the domain the content will still be reachable
<glowplug>
The wiki is likely sufficient anyways. What if I designed the page anyways then we can link to the wiki. Thats a fairly common way of doing things now adays I think. =)
<azonenberg>
I just want there to be one central place for all of the info
<glowplug>
Pretty front page. Then link to the wiki for community documentation ect.
<azonenberg>
i dont want a collection of a dozen different websites
<azonenberg>
Design it, but don't put it anywhere public just yet
<glowplug>
Haha
<glowplug>
It wont be that bad! O_O
<azonenberg>
I just mean i dont want it showing up in search results before we have anything on the back end
<glowplug>
The mediawiki page is actually a great idea. It's relatively easy to add content to a regular HTML5 page but the wiki format is a LOT easier. Also easier to keep track of permissions ect.
<azonenberg>
Yeah
<azonenberg>
Then we can have a github or something for CAD files and stuff
<glowplug>
I was actually mad for a second when I noticed the google page wasn't updated since 2011. I literally said outloud. "There is no fucking way that they gave up, its too important!" Haha
<azonenberg>
I did not give up, i just got really busy
<azonenberg>
grad school happened :p
<glowplug>
Biggest relief ever...
<azonenberg>
So it's been back-burnered but is very high on my "things i want to do" list
<glowplug>
Almost had a heart attack at 26.
<azonenberg>
I just want to be able to build a board with an ASIC of my own design on it :D
<azonenberg>
Also, re packaging
<glowplug>
And even though they don't know it yet.
<glowplug>
That's what everyone else on Earth needs.
<azonenberg>
My tentative thought is to solder-bump the bare die and make a flip-chip CSBGA
<azonenberg>
Wire bonding is a huge pain in the neck
<azonenberg>
but if i could just put down a thick overglass layer, etch contact windows, then put some metal down
<azonenberg>
then manually place a solder ball over each bond pad and reflow
<glowplug>
Flip chip is pretty much a neccessity in my opinion. At 405nm the only way to get performance that people will "notice" is with very high frequencies, which means high TDP.
<azonenberg>
That too, but that's longer term
<azonenberg>
I'm mostly thinking what's realistic for a homebrew lab
<azonenberg>
you want to minimize use of complex equipment
<azonenberg>
For example, PECVD needs all kinds of nasty gases
<azonenberg>
But reactive sputtering can be done with just argon and O2/N2
<azonenberg>
Ar + O2 on a Si target = SiO2
<glowplug>
Flip chip style packages are feasable for homebrew no?
<azonenberg>
I think they're more feasible than wire bonding
<glowplug>
Then I think we win. 8)
<azonenberg>
wire bonding is a huge pain
<azonenberg>
i've done it
<azonenberg>
or tried, anyway
<azonenberg>
in a real cleanroom
<glowplug>
Haha
<azonenberg>
several hours got one bond :p
<azonenberg>
one end of one, that is
<glowplug>
That.. does not sound like fun...
<azonenberg>
Constantly re-threading the tip with tiny tweezers
<azonenberg>
when the gold wire kinked or broke
<azonenberg>
It was not fun
<azonenberg>
I asked to get trained on it
<azonenberg>
the first thing the guy did was try to talk me out of it
<azonenberg>
then said "if you're sure, go for it"
<glowplug>
To expand on what you mentioned before with process sizing. I think that building very LARGE devices until they work correctly might be a great way to go about it. Like 500um transistors and logic gates.
<azonenberg>
500um is a little big
<azonenberg>
i'd work at like 50
<glowplug>
Whatever sizing is extremely comfortable basically.
<azonenberg>
But meanwhile, try to get the litho working better much smaller
<glowplug>
That way we know the issue is in the design and not the implimentation.
<azonenberg>
Yeah
<azonenberg>
But the first step is to get working litho
<azonenberg>
Then we need to get metallization figured out
<azonenberg>
that's a problem jeri hasnt solved
<azonenberg>
I'm debating Al or Cu
qi-bot has quit [Ping timeout: 245 seconds]
<azonenberg>
I was thinking copper damascene might be cool to do
<glowplug>
I built my CNC for $400 and I know that I can get 50um accuracy. Blu-ray diodes are cheap. The decently accurate litho I think is fairly easy.
<azonenberg>
but i dont know if aluminum might be easier
<glowplug>
This is the doping or empregnating process correct?
<azonenberg>
No
<azonenberg>
Doping is a mostly solved problem
<azonenberg>
Emulsitone sells spin-on glass with dopants dissolved init
<azonenberg>
in it*
<azonenberg>
Coat undoped glass, etch windows where you want to dope
<azonenberg>
coat doped glass
<azonenberg>
diffusion bake
<azonenberg>
strip glass in HF
<azonenberg>
Jeri did something similar and it worked well
<glowplug>
Those dopants use known or proprietary chemical formulas?
<azonenberg>
Mostly known, though the exact details are proprietary
<azonenberg>
Getting that kind of purity in a home lab wont be easy, better off to start with commercial stuff
<glowplug>
Using proprietary dopants is better than proprietary semiconductors, but not much better.
<azonenberg>
Well, they're known to be boron and phosphorus ion sources
<glowplug>
Thats good enough for me. At least reproducing them is theoretically possible.
<glowplug>
VTR is an implimentation of a theoretical FPGA architecture.
<azonenberg>
I should look at it and see if it can be ported to a real FPGA
<glowplug>
Creating even an absolutely basic circuit, lets say 5-10 LE's that is compatible with that toolchain would be game-changing.
<azonenberg>
Let's make a 74HC series chip first lol
<glowplug>
It may even be the fastest way for this project to get noticed.
<azonenberg>
THAT would get noticed
<azonenberg>
flip-chip BGA 74HC04
<glowplug>
This could be used to build an entire "computer" system correct?
<glowplug>
Oh that is a hex inverter.
<glowplug>
Interesting.
<azonenberg>
Yes
<azonenberg>
But nobody has done it yet
<glowplug>
It's pretty easy to see off into the future because I wasn't around for the 70s.
<azonenberg>
Lol
<glowplug>
If we could produce the entire 7400 line of ic's and build a basic computer. That would infact be revolutionary
<glowplug>
Not the entire line. But enough to make a board that can load and execute programs, read and store memory ect.
<azonenberg>
One thing at a time
<azonenberg>
Get litho working :p
<glowplug>
So I strap a blu-ray diode onto my CNC. What other equipment do I need to do something actually useful. Even if I can do it then ship it to someone thats fine.
<azonenberg>
Well you'd need clean enough air that you can make stuff without too many defects
<azonenberg>
so a wood shop etc is a bad location
<azonenberg>
a desk in a living room is probably enough for reasonable sized stuff t otest
<glowplug>
The area I'm in is fairly clean.
<glowplug>
I need photoresist thats pretty obvious.
<glowplug>
On the google page it says laser-direct litho for mask fabrication.
<azonenberg>
That was the plan, yes
<soul-d>
my brain is still locked in some infinite limit loop... tried to open a calculus book this week and it broke stuff :(
<glowplug>
Ahah!
<glowplug>
I was under the assumption that you were burning away photoresist with lasers directly.
<azonenberg>
No
<azonenberg>
it's for exposure
<azonenberg>
so you need much lower powers
<azonenberg>
Otherwise why bother with photoresist
<azonenberg>
just do ablation of a hardmask
<glowplug>
Have you considered that aproach?
<azonenberg>
Yes, you need higher laser powers and risk thermal damage
<glowplug>
So high cost and low yields.
<glowplug>
Alright so really we need masks made.
<azonenberg>
The main reason being that you cannot easily align a laser system to an existing pattern
<glowplug>
What is the mask material?
<azonenberg>
but aligning a mask is pretty straightforward
<azonenberg>
Anything opaque and etchable
<azonenberg>
on a glass plate
<azonenberg>
evaporated metal is a good choice to start
<azonenberg>
For the first step, though, focus on patterning photoresist
<azonenberg>
on anything
<glowplug>
Using the masks correct?
<azonenberg>
Direct laser on photoresist
<glowplug>
What thickness metal film do I need?
<azonenberg>
Oh, nanometers is fine
<azonenberg>
half a micron at most
<glowplug>
Alright so the Mask is .5um and I handle it between two pieces of glass?
<glowplug>
I cut it out with the blu-ray laser...
<glowplug>
Why practice with directly burning the photoresist?
<azonenberg>
Not exactly
<azonenberg>
The normal mask structure would be a film of metal deposited on a single layer of glass
<azonenberg>
with the metal exposed
<azonenberg>
You'd coat photoresist on the metal
<azonenberg>
expose that with the laser, develop
<azonenberg>
then etch the metal and strip the resist
<azonenberg>
now you have a metal-on-glass mask
<glowplug>
Alright so this is exactly the same thing as making a PCB except exposure is with a laser diode instead of UV light. And the "pcb" is a .5 micron thin sheet of metal on a piece of glass.
<glowplug>
The entire thing is the mask.
<azonenberg>
Yeah
<glowplug>
Alright thats not too bad...
<soul-d>
so steps azon described is the iron on step for pcb's :P
<azonenberg>
Then to use it, you flip it over so the metal is touching the wafer
<glowplug>
I'm trying to find the metal film. Do you have any links on hand?
<azonenberg>
align, expose, andd then do normal litho on the wafer
<azonenberg>
You don't buy film
<azonenberg>
you buy glass and coat it yourself with an evaporator or sputtering system
<azonenberg>
Building that is one of the to-dos
<azonenberg>
but it's nontrivial
<azonenberg>
Which is why i suggested bringing up the lithography first
<azonenberg>
with photoresist on bare glass
<glowplug>
Oh shit in a shit basket.
<glowplug>
Alright I'm starting to grasp the process now.
<glowplug>
So if I can etch photoresist on bare glass with decent accuracy then doing so to protect a layer of .5 micron metal is a small step.
<glowplug>
Sounds good I'll see what I can do. Hopefully you dont get an evaporator too much faster than I master the litho. =P
<azonenberg>
Once you can coat the metal at all
<azonenberg>
I have access to an evapoarator on campus now that i can use any time i want
<azonenberg>
i just dont have one of my own
<azonenberg>
So i can coat a bunch of microscope slides etc relatively easily
<glowplug>
So just to double check that I understand perfectly.
<glowplug>
My end product should be a piece of glass with a detailed pattern of undamaged photoresist material correct?
<azonenberg>
For the first test, yes
<glowplug>
Alright kick ass.
<azonenberg>
Do you have a microscope sufficient to inspect it?
<azonenberg>
And the substrate doesn't even have to be glass
<azonenberg>
it could be anything flat
<glowplug>
Not here. But I think i3 Detroit has one.
<azonenberg>
metal might even be easier
<azonenberg>
lik a PCB or something
<glowplug>
If not then I can try to acquire one.
<azonenberg>
glass may require an adhesion promoter to stick well
<glowplug>
I will try ~50 micron patterns. So I need a 50x microscope?
<azonenberg>
That'd probably work fine
<glowplug>
There is one person trying to decode logic gate devices?
<azonenberg>
Me and john mcmaster are the folks behind siliconpr0n
<azonenberg>
if thats what you mean
<glowplug>
I mean super basic transistor level gates NAND ect.
<glowplug>
Without those designs its impossible to make "actual" useful devices.
<azonenberg>
Oh
<azonenberg>
Once you can make transistors, building gates is trivial
<glowplug>
Well... as long as its trivial. Haha
<azonenberg>
i have a couple of simple gates drawn out already (though not simulated or tested)
<azonenberg>
i mean, a NOT gate is literally an NMOS and a PMOS in series
<azonenberg>
with the gates connected together, low side of the NMOS to ground, high side of the PMOS to power
<azonenberg>
and output at the middle
<glowplug>
I will google it up then and get a design into InkScape then use that for the litho practice.
<azonenberg>
I advise against that
<azonenberg>
For litho practice you want a repeatable pattern like a grid or parallel lines
<azonenberg>
so it's easy to measure process variation
<glowplug>
Ahh good point.
<azonenberg>
A single layer of a logic gate mask is just going to be a rectangle or two
<azonenberg>
not enough to get any real data from
<glowplug>
It's hard to verify repeatability of my machine with complicated designs.
<glowplug>
Fair enough. I'll make a grid. =(
<azonenberg>
Another good test pattern would be a set of parallel lines of gradually decreasing width
<azonenberg>
See how small you can resolve them
<glowplug>
Alright I will do that too.
<glowplug>
I'm still absolutely blown away that more are not working on this.
<azonenberg>
Most people just assume it's impossible or too hard
<glowplug>
I'm pretty sure that I'm not insane in understanding how fundamental this is to the future of technology.
<glowplug>
At any rate if we can get a working 50 micron transistor I have a feeling the amount of contributors will be quite a lot.
<glowplug>
Then again maybe we are insane. Who knows. O.o
<glowplug>
What is the light source for exposing the photoresist? Standard UV?
<azonenberg>
The stuff i've been using is wide-band
<azonenberg>
peak sensitivity around 405nm
<azonenberg>
which just happens to be the center frequency of a blu-ray diode
<glowplug>
We wont be using lasers to expose through the mask though correct?
<azonenberg>
No
<azonenberg>
You can use a regular blacklight bulb from a party store
<glowplug>
I have 3w UV LED's that I use for exposing photoresist for PCB's. They are probably bright enough.
<azonenberg>
Yeah
<azonenberg>
I usually use thinner layers for microfab than PCBs so the exposure time might even be less
<glowplug>
I think they are 395nm.
<azonenberg>
That'll work too
<azonenberg>
350 - 450 i think is what the injectorall resist is sensitive in
<glowplug>
So I can use the negative photoresist sheets that I have for testing?
<azonenberg>
Sure, the exact resist is unimportant for that step
<glowplug>
Well that is pretty much badass.
<glowplug>
So I just need a blu-ray diode... literally thats it.
<azonenberg>
What matters most is to demonstrate direct-write exposure at fairly small scales
<glowplug>
OH right and 50 micron accuracy...
<glowplug>
Thats .05mm. Yeah I can't see any reason why my machine couldn't do that.
<glowplug>
I will see if I can get 50x photographs of my progress.
<glowplug>
That shouldn't be super difficult either.
<azonenberg>
If you get something that looks good and can package it safely i can give you my office address to mail it to
<azonenberg>
and i can do higher res imaging
<azonenberg>
Keep me posted :)
<glowplug>
Absolutely. I can just hang out in the channel I'm usually in the opencores channel all the time too.
<glowplug>
Let me know when you get the wiki page up.
<azonenberg>
Bug me tomorrow and i'll probably do it then
<azonenberg>
kinda busy right now
<glowplug>
Haha
<glowplug>
Alright I'll do that. Trying not to get overly excited since this is a painful slow incrimental process. But it is seriously a big deal. I hope more people start to notice. =)
soul-d has quit [Remote host closed the connection]
superkuh has quit [Quit: DayZ]
Prottey has joined #homecmos
Prottey has quit [Remote host closed the connection]
Prottey has joined #homecmos
superkuh has joined #homecmos
soul-d has joined #homecmos
<Sync__>
building an evap is not that hard
<Sync__>
building a MBE is hard
qi-bot has joined #homecmos
<Sync__>
oh haha azonenberg they use sparkplugs. m)
B0101 has joined #homecmos
<glowplug>
Molecular beam epitaxy? Do we need such a thing for +405nm sized devices?
<Sync__>
no but it is nice to have
<glowplug>
Thank goodness. Haha
<Sync__>
most of my work has to do with mbe
<glowplug>
I see.
<Sync__>
but it wouldn't be too hard to make a system
<glowplug>
I think that if we do get DIY MBE it will also come with superior litho/optics together as a package. That is probably very far in the future.
<Sync__>
if you are lucky maybe 10k for everything
<glowplug>
I would say mostly for cost reasons.
<glowplug>
^Exactly. Haha
<Sync__>
10k is not a lot
<Sync__>
the more annoying things then is the analysis hardware
<Sync__>
I mean, it's a hobby, that's where you spend your money
<glowplug>
I agree. But people are still trying to get CNC's and 3d printers down below $400 each. I built my CNC for only $400. I think sub $1k for all equipment for an entire lab is the pricepoint we need to get a true DIY following.
<Sync__>
I don't get why people even bother with those shabby 3d printers
<Sync__>
or those clapped out cncs
<glowplug>
That pricepoint isn't necessary for progress. Obviously not. But it will get a real massive following of people to explode research and resources.
<glowplug>
My CNC is fairly accurate (50 microns). The 3D printers I mostly agree with your point. The usefulness of plastic parts of that accuracy is not super great.
<Sync__>
50 microns is not accurate imho
<Sync__>
I expect a cnc mill to be able to produce bores tolerated close enough for bearings ~j5-6
<glowplug>
In the grand scheme of things no its not. But its accurate enough to make 50um masks for the litho process. Yes 50 micron semicunductors are not super useful. But its good enough for research and to prove concepts before scaling down.
<glowplug>
Basically I agree with you on all points in the grand scheme. But there are some major consessions that have to be made in the name of DIY, cost, and getting the word out to expirimentors.
<glowplug>
Then when the resources and peoplepower allow. We can get "real" equipment. =)
<glowplug>
Be right back have to go to the post.
<Sync__>
I'm probably spoiled by having too much money to spend
<Sync__>
but yeah it's an interesting problem to get sub µm positioning accuracy when you have a mechanical stage and piezos are out of question
<Sync__>
I guess one could try preloaded ballscrews and two interferometers
<Sync__>
even 5 micron positioning would be easy enough to do for a hobby machine
<soul-d>
actualy
<soul-d>
3d printer could be handy if the parts are acurate enough to be used as mold for say aluminum /metal or alloy
<Sync__>
yes, professional 3d printers are able to do that
<Sync__>
or at least provide good enough finish to be used to laminate off them
<soul-d>
although bit redundant might as well cnc them directly ( i assumed most home cnc's don't have the strengt or accuracy for metal ) but still some parts might be better off with a mold
<azonenberg>
Sync__: i was thinking of making ~5um features photomasks with direct write litho
<azonenberg>
followed by ~5x optical shrinking
<azonenberg>
for ~1um features
<Sync__>
yeah
<Sync__>
that'd be easy
<Sync__>
the only issue would be the cumulative error if you do not have an interferometer or linear encoder
<azonenberg>
Yeah, you'd need feedback for sure
<azonenberg>
You can get encoder strips on plastic film from laserlab at 12.5um half-pitch (25um pitch)
<azonenberg>
then use interferometry for reading the position between lines
<Sync__>
that's too much hassle
<Sync__>
you'd need to develop all that
<azonenberg>
Other option is a rotary encoder geared down
<Sync__>
I'd just get chines scales
<Sync__>
well that does not help the cumulative error
<Sync__>
which you will always have
<azonenberg>
What do you suggest for that?
<Sync__>
because of pitch error
B0101 has quit [Read error: Connection timed out]
<Sync__>
well, for the closed loop an encoder on the screw and to get the pitch error out cheap linear scales
<azonenberg>
PItch error is probably fine
<azonenberg>
as long as you dont have backlash and it's repeatable
<azonenberg>
right?
<Sync__>
the regular positioning scheme for everything cnc
<Sync__>
well, usually screws are specced for 1/100mm per 300mm
<azonenberg>
10um absolute error
<azonenberg>
that's not 10um of jitter
<Sync__>
that's not even accounted for
<azonenberg>
also, 300mm is HUGE unless you're making something like a full 8" wafer
<Sync__>
because it could jitter around in the 300mm
<azonenberg>
let's say you're using a 40mm wafer
<azonenberg>
so positioning accuracy of 1.3 microns across the entire wafer?
<azonenberg>
With repeatability quite a bit better than that>
<azonenberg>
?
<azonenberg>
I'd take that
<Sync__>
if you trust the screw in the range you use
<Sync__>
but you could just check that with an indicator
B0101 has joined #homecmos
<glowplug>
Ok theres a lot of information here.
<glowplug>
Firstly this page has everything you need to know about making accurate plastic parts.
<glowplug>
Secondly the CNC / encoder accuracy. It's not something I have physically at the moment but something that I've solved and am working on.
<glowplug>
My next CNC will have closed loop 200,000 steps per revolution bldc motors (you could call them servos).
<glowplug>
It is possible to position the machine very accurately using a conductor the diameter of a hair and a conductor around the perimeter of the part. When they make contact you get a zero position. That obviously has limitations but its probably good enough. Inteferometry is easy to build but hard to measure.
<glowplug>
Also I don't think that I can get better than 10 microns on my next machine. I hope that is accurate enough. :/
<glowplug>
Actually I may have spoke too soon. I found someone just now with a DIY CNC who is getting 1 micron accuracy engraving with a diamond tip. Interesting.
<Sync__>
encoder resolution is not really everything :D
<Sync__>
because mostly the mechanical accuracy is far inferior
<glowplug>
Agreed. I have been researching this problem over the last month or so. Unfortunately it did turn out that encoder/motor accuracy was the easiest problem to solve.
<glowplug>
That machine is what I need to aproximate. Unfortunately it costs $20k.
<glowplug>
Repeat accuracy .05mm under no load (blu-ray laser in my case).
<glowplug>
Wait thats 50 microns not 5. Hmm.
<glowplug>
Michal (the google employee who wrote the article I linked above) gets ~2 micron accuracy with that mill. I wonder how.
<glowplug>
It has a state mechanical resolution of .001mm but that is not repeat accuracy (clearly).
<Sync__>
wow they actually want 20k for that
<glowplug>
That repeat accuracy is probably for very large parts.
<glowplug>
Here is a qoute from Michal.
<glowplug>
""But in the end, I ended up with accuracy routinely in the vicinity of 0.002 mm for small parts."
<Sync__>
in the material that is not a problem
<glowplug>
Also he is milling at .002mm. I will be using a laser.
<Sync__>
milling is just fine
<glowplug>
It just means that in theory I should have an easier time getting the same repeat accuracy.
<Sync__>
I'd just get a few surplus linear slides preferrably preloaded ones and new hiwin preloaded ballscrews
<Sync__>
then it should be no issue at all to get that level of positioning accuracy
<glowplug>
Also his machine has a massive work area. I only need 2 micron accuracy in a square inch workspace at most.
Prottey has quit [Quit: Leaving.]
<glowplug>
I'm researching the chinese ballscrews/nuts right now. They may be accurate enough at $77 per axis.
B0101 has quit [Ping timeout: 245 seconds]
<azonenberg>
glowplug: yeah, it gets much easier to get accuracy across a small area
<azonenberg>
Especially if you restrict the prototype to single dies rather than a full wafer
<azonenberg>
in industry people are trying to work with 18" wafers
<azonenberg>
rather hard by comparison :p
<glowplug>
Absolutely. And that kind of scaling is an economical consideration for production. That is very very very far in the future.
<glowplug>
I think some transistors and logic gates would be a massive feat. =)
<glowplug>
It looks like I'm going to go with chinese ballscrews and a double ballnut per axis configuration.
<Sync__>
azonenberg: interestingly those large wafers rely on impurities from the melt for stability
<glowplug>
Should be ~$300 per axis.
<Sync__>
huh double ballnut is eh
<azonenberg>
glowplug: for hobbyist volume 2" wafers is more than enough
<Sync__>
get real preloaded ballscrews
<glowplug>
I had a very strong feeling you were going to say that.
<Sync__>
the cost of the double nut is not worth it
<glowplug>
I'm on a pretty limited budget. If I can get 2 micron repeatable accuracy with a $500 CNC I will be happy. If we need more accuracy than that then I need a grant / donation. Haha
<Sync__>
I'm not even suggesting planetary screws!
<glowplug>
The ballscrew/nut sets you are suggesting cost per axis the same as my entire planned machine. =P
<Sync__>
no
<glowplug>
Do you have any links?
<Sync__>
price out some hiwin spindles
<glowplug>
Those are the sets I'm talking about.
<glowplug>
They are $300+
<glowplug>
Versus $100.
<glowplug>
Hiwin KK10020P40238 is $460 per axis.
<Sync__>
wat
<Sync__>
well no shit
<Sync__>
you are looking at ground ones
<Sync__>
also much too large ones
<glowplug>
Thats why I asked if you had links.
<glowplug>
Hiwin Tech 9812963004 are $490 each.
<Sync__>
I wrote them an email a few years ago
<Sync__>
and requested some pricing
<Sync__>
and it was way less
<glowplug>
Alright I did find something.
<glowplug>
Nevermind those are also chinese.
<glowplug>
Yeah I don't know. All I can find are surplus Hiwin and those are still $400+
<Sync__>
if you use only positive approaches backlash does not matter
<Sync__>
so you could just use acme screws
<glowplug>
You mean never changing direction while running the laser?
<Sync__>
basically
<Sync__>
you you only laser while moving forward
<glowplug>
What happens when the machine changes direction after passing over the part to make another pass?
<glowplug>
Wouldn't it lose absolute accuracy there?
<Sync__>
you get backlash
<glowplug>
But its known backlash that does not damage the part. So you compensate in software?
<Sync__>
well if it is always the same you can compensate it in software
<glowplug>
Is it common for the backlash to be always the same? Are there digital tools for measuring backlash during operation? I have never even researched that.
<Sync__>
you can use a dti for measuring backlash
<glowplug>
Looks like indicators with USB are ~$300.
<glowplug>
I will try to track down something cheaper.
<Sync__>
why do you want usb
<Sync__>
just read it
<glowplug>
Because EMC2 has to make the adjustments. I'm not making them by hand every pass screw that. Haha
<Sync__>
I do have an extramess with a resolution of 200nm
<Sync__>
oh
<Sync__>
you did not understand me
<Sync__>
you can just measure it one time and set it
<Sync__>
backlash will be the same
<Sync__>
if you want that get linear encoders
<Sync__>
linuxcnc can do rs422
<glowplug>
Backlash will be the same every time no matter what?
<Sync__>
mostly
<Sync__>
but you can measure that
<glowplug>
So I would have to measure every pass?
<Sync__>
no
<Sync__>
just let the spindle run a few passes measure average backlash and put that value in emc
<glowplug>
And you think that average backlash is sufficient for 2 micron accuracy?
<Sync__>
you gotta try
<glowplug>
That is deffinately the easiest way. If it fails however.
<glowplug>
Couldn't I take constant measurements with a digital indicator and apply that in realtime for backlash compensation?
<Sync__>
there are linear encoders for that
<glowplug>
What do you think about using a combination of rotary and linear encoders?
<Sync__>
that ist the standard solution for cnc drives
<Sync__>
+t
<Sync__>
gah -t
<glowplug>
Alrighty. Looks like thats what I'll do then.
<Sync__>
well 1µ glass slides are not exactly cheap
<Sync__>
still affordable
<glowplug>
Yeah around $350. I'm going to see if I can DIY something.
<Sync__>
that's not worth the effort
<Sync__>
those are things that are just not worth making
<glowplug>
Unless it works. Then it was worth it. Haha
<Sync__>
there is just too much hassle involved
<Sync__>
if one calculates the hours you put in it just does not make sense