#homecmos IRC log for Sunday, 2011-08-07

azonenbergJust finished a test of Cu patterning03:38
azonenbergComplete success03:38
azonenbergPics uploading now, still writing up lab notes03:39
wolfspraulsuccess sounds good!03:40
wolfspraulkristianpaul and I added an entry about your work in the news we will push out Monday http://en.qi-hardware.com/wiki/Copyleft_Hardware_News_2011-08-08#Homebrew_CMOS_and_MEMS_foundry03:41
azonenbergAnother 20um feature size nyan cat03:41
azonenbergI have a much better photo for you to use03:41
azonenbergstill uploading03:41
azonenbergThe ones you have were overetched03:42
azonenberguploading to http://colossus.cs.rpi.edu/~azonenberg/images/homecmos/2011-08-06/03:43
azonenbergi'll send exact links when they're done03:43
wolfspraulI confess that I am mirroring your entire images/homecmos folder :-)03:43
azonenbergLol be my guest03:43
wolfspraulbecause it's so good, want to be able to browse you quickly with my feh viewer03:43
wolfspraulso I can show people and browse them through your greatness03:43
wolfspraulit's good work03:44
azonenbergtoday's stuff has another minute left to upload then you can grab it03:44
azonenbergi'll narrate when its done03:44
Action: azonenberg finishes writing lab notes03:44
wolfspraulyes, and that too03:44
wolfspraulvery methodical notes, good pictures. it's a joy.03:44
wolfspraulnow we only need to find a way to bring this into use cases...03:44
wolfspraulwhich should slowly emerge out of it, somewhere03:45
azonenbergI'll have a comb drive soon, if luck is with me03:45
azonenbergmost of the big problems are solved03:45
azonenberg-- photos done uploading03:45
CIA-67homecmos r107 | trunk/lithography-tests/labnotes/azonenberg_labnotes.txt | Today's lab notes - moar nyanotechnology!03:48
azonenberghttp://colossus.cs.rpi.edu/~azonenberg/images/homecmos/2011-08-06/S7301594.JPG and 1595 were my first attempt after developing but before etch03:50
azonenbergmag is 100 and 400x respectively, still 20um pixels03:50
azonenberg1596 and 1597 were after etching (not long enough)03:50
wolfspraulthe problem is I also have the toped screenshot03:50
azonenberg1598 is after a proper etch03:50
wolfspraulif I change the pic it won't match anymore, would be misleading?03:51
azonenberg1599 is 40x, 1600 is blurry, 1601 is 400x03:51
wolfspraulor is that screenshot still accurate?03:51
azonenbergI used the same mask03:51
azonenbergthis was just another exposure off it03:51
azonenbergusing different etch parameters03:51
wolfspraulah ok03:51
azonenbergsame size, too03:51
azonenbergthen 1602 is 40x after stripping photoresist03:51
wolfspraulbtw, you say picel size 20um, and then you say the cat is 'about 200 um'03:51
wolfspraulbut I count many more than 10 rows, almost 2003:52
azonenberg1603 is 100x, 1604 is 400x03:52
azonenbergAnd let me double check my measurements03:52
wolfspraulyou say the cat is 600 um long, 200 um high03:52
wolfspraulbut I think the height is more like 300 um03:52
azonenbergYeah, i think you're right03:52
azonenbergthat was probably a typo on my part03:52
wolfspraulwell. in all my stupidity, all I can do is to count the rows, and multiply by 20 :-)03:53
azonenbergi count 18 rows so thats 18*20 = 360um03:53
wolfspraulbut I guess at least that I did right...03:53
azonenbergso not usre how i got 20003:53
azonenbergIn any case i'd appreciate you replacing the overetched pics with ones from this die03:53
azonenbergyou can keep the screenshot03:53
wolfspraulwill do03:53
wolfspraulI have about 500-800 readers03:53
azonenbergNice :)03:54
wolfsprauland it's buried somewhere in a lot of stuff, but some people will see it I think03:54
azonenbergCant hurt to get the word out03:54
wolfspraulmaybe I bump it to the top03:54
R0b0t1OMFG NYAN!03:55
azonenbergR0b0t1: Nyanotechnology03:56
azonenberghttp://colossus.cs.rpi.edu/~azonenberg/images/homecmos/2011-08-06/S7301603.JPG is a 100x view of the finished chip03:56
azonenbergeach of the pixels are 20 microns (0.02 mm) across03:56
wolfspraulazonenberg: so which pictures would you propose to use? I think two is fine, one zoom out (entire cat), one zoom to just the head03:56
wolfspraulwhich ones are most representative of your work?03:56
azonenbergI'd say 1603 and 160403:57
azonenberg100x and 400x mag respectively03:57
R0b0t1I'm glad to see it looked like it finally worked03:57
azonenbergThis is my best work to date re metal layer patterning03:57
azonenbergR0b0t1: Yes, it worked beautifully03:57
azonenbergConveniently Cu is also HF resistant03:57
azonenbergSo i can use it as a hardmask for patterning SiO2 or Ta2O503:57
wolfspraulthe cat is copper on silicon?03:57
azonenbergBlue is Si, orange is Cu03:58
azonenbergevaporated around 200nm Cu on the whole die and then etched down to make the lines03:58
wolfspraulazonenberg: updated http://en.qi-hardware.com/wiki/Copyleft_Hardware_News_2011-08-08#Homebrew_CMOS_and_MEMS_foundry04:29
azonenbergLooks great, thx04:30
wolfspraulthank you, inspiring work, keep it up04:30
wolfspraulis 4004 CPU still on your horizon? or something else now?04:30
azonenbergIts one of the several relatively large scale (1E3 transistor range) devices i'd like to build in the long term04:31
azonenbergNow that metal etching is solved and Si etching is coming along nicely, i think i am very close to a comb drive04:31
azonenbergwhich is the immediate goal04:31
wolfspraulI'm just trying to understand your mental roadmap, and I fully understand it may change as you move forward and discover more, successes and failures etc.04:31
wolfspraulwhat is a comb drive?04:32
wolfspraulI am seriously interesting in finding a way to embed some of your stuff on hacker boards or even products I could build04:32
wolfspraulnot now of course, I understand there's more work to be done first04:32
azonenbergA comb drive is a capacitive linear actuator04:33
azonenbergLet me upload a GDS and i'll walk you through it h/o04:33
wolfspraulI think that would be so cool if we could squeeze out some actual use case out of what you build, from the perspective of a software developer04:33
wolfspraulno matter how small the 'programmability' may be at the beginning04:33
wolfsprauljust explaining my perspective04:33
azonenberglook at the cell TestDie104:35
azonenbergIn the background we have the backside etch layer04:36
wolfspraulhe, I need to install toped first, in fact build from source04:36
wolfspraullet me try...04:36
azonenbergkk, let me know when you have it open04:36
azonenbergSo the background layer is the backside etch04:52
wolfspraulthose 2 big blocks at the top and bottom?04:52
azonenbergNo, zoom out further04:52
azonenbergits a ring around the whole thing04:52
wolfspraulI'm zoomed out I think04:53
azonenbergThe idea here is to thin out the central part of the die by etching from the back towards the front04:53
wolfspraulI see the text on top and bottom04:53
azonenbergNo, its beyond that04:53
wolfspraul"comb drive xx.1" and "2011.x.2."04:53
wolfspraulfurther out?04:53
azonenberglike by 4x04:53
wolfspraulok there are 2 squares around it04:54
azonenbergThe outer ring is where we keep the wafer at full thickness04:54
azonenbergEverything inside there is thinned down to around 20-30 microns04:54
azonenbergthis is done with KOH from the back toward the front04:54
azonenbergThe idea is to leave a nice stiff carrier around the edges so we can pick the die up without cracking it04:55
wolfspraulwhen you say 'ring' you mean the space between the 2 outer squares, right?04:55
azonenbergThe outr square04:55
wolfspraulsure I was just confused because when I see 'ring' I think 'round'04:55
azonenbergIts a square ring04:55
azonenbergturn on shading for all of the alyers04:55
wolfspraulI got it04:55
azonenbergAnyway so thats the backside etch04:55
azonenbergOnce we have the wafer thinned out, we coat the wafer with hardmask and etch the red mask04:56
azonenbergsorry i forgot gds doesnt have colors lol04:56
azonenbergits the layer that doesnt have text on it04:56
azonenberg1 i think04:56
wolfspraulyes, no colors right now04:56
wolfspraulin my view04:57
azonenbergThis is the active area mask04:57
wolfspraulthinking ahead, how do we cut this out of the wafer and make any i/o pins (power, gnd, data) accessible to the circuit around the chip?04:57
azonenbergThe entire background should be protected (I just didnt draw it filled since the mask isnt finished)04:57
wolfspraulin other words - how do we package or seal the chip?04:57
azonenbergEtch this down until we've penetrated the entire die (fingers hang free)04:57
wolfspraulthe active area is in the middle I guess04:57
azonenbergthen deposit metal and etch the mask with text on it04:57
azonenbergSo at this point we have two sets of fingers04:58
azonenbergone mounted on a spring and one hanging free04:58
azonenberg*one mounted on a spring and one attached to the body04:58
azonenbergThey're both covered in metal but are isolated from each other electrically04:58
wolfspraulyou are talking about packaging now?04:58
azonenbergstill how it works04:59
azonenbergWhen we apply a voltage across the top and bottom areas, the plates develop a charge and attract04:59
wolfspraulah ok04:59
azonenbergso the springy fingers move toward the stationary ones04:59
wolfspraulwait, I zoom in :-)04:59
azonenbergthen when we remove the voltage they spring back04:59
azonenbergThe design here is still early, it needs a lot of tweaking04:59
azonenbergAnyway so this is meant to be used (in the end) as part of a system with moving parts of some sort05:00
azonenbergits basically a linear motor05:00
azonenbergRe packaging, for ICs (rather than MEMS) since they have no moving parts, its relatively easy05:00
wolfspraulit's this I guess... http://en.wikipedia.org/wiki/Comb_drive05:00
azonenbergYou'd make the die a mm or two across and scribe it out of the wafer (I separate my dies from the wafer before processing since i only make one die at a time)05:01
azonenbergThe top metal layer would be extra thick05:01
azonenbergand then you could wire bond or similar to it (detials of how to attach wires are TBD)05:01
azonenbergbut you'd have 50-100 micron wide contact pads around the rim of the chip05:01
azonenbergthen you seal it on with a blob of epoxy05:01
wolfspraulis the packaging still part of your lab setup/goal, or would you use a service for that?05:02
wolfspraulwirebond dies typically come in small chocolate-sized vacuum sealed packlets05:02
azonenbergThe immediate goal is fab, i'll be testing them with microprobes05:02
azonenbergif and when i build something suitable for actual use in a project i'd think about it then05:03
azonenbergI might be buying/building a wirebonder eventually05:03
wolfspraulthen it goes to a wirebond machine to run the wires (aluminum, gold, etc) and put the epoxy on as a last step05:03
wolfspraulall of which is done after smt/reflow because it cannot take heat05:03
azonenbergInteresting, i didnt know that part05:03
wolfspraulof course you know that, I'm just explaining where my current knowledge stops05:03
azonenbergthe epoxy cant handle reflow?05:04
wolfspraulI can double-check, haven't done wirebonding for a year or so05:04
azonenbergBecause the exact same process (excpet the epoxy is in a chip-shaped mold and has a leadframe around it instead of a PCB) is used for making normal ICs05:04
wolfspraulthere are wire-bond modules which in turn go through reflow05:04
azonenbergthey wirebond from the die to the leadframe then fill the whole thing with epoxy05:04
wolfspraulmaybe some can and some cannot05:04
azonenbergchip on board simply removes some of the middleman05:04
wolfspraulbut also adds new ones05:05
azonenbergnot really05:05
azonenbergInstead of a separate soldered leadframe you bond from the die straight to the copper traces05:05
wolfspraulin the industry, many smt shops do not have a wirebond machine05:05
azonenbergthen epoxy fill over it05:05
wolfspraulfor some strange reason05:05
azonenbergI want one :P05:05
azonenbergi've used them in the cleanroom05:05
wolfspraulso if you have a product with wirebond, the logistics and testing has more steps05:05
wolfspraulbecause stuff is sent around05:05
wolfsprauloh it's no rocket science for sure, I'm just saying it's a separate/different workstep and as such adds complexity to the manufacturing process05:06
wolfspraulwhich if you have stronger packages, someone else has done this before05:06
wolfspraulthe stronger packages come at a price too of course05:06
wolfspraulmost wirebond places are small offices, like a dentist05:06
azonenbergLike i said i havent done much COB (except for taking them apart)05:07
wolfspraulI've heard that some smt fabs now add this machine (and workstep) to their arsenal, but I think for the most part it's still separate05:07
azonenbergi've only studied wirebonding as it applies to fabrication of regular ICs05:07
wolfspraulwirebond is definitely disruptive to the manufacturing and testing process05:07
azonenbergor for bonding a prototype die for packaging05:07
wolfspraulso that additional disruption (cost) needs to be made back, which is why you find wirebond typically only in high-volume applications (say 100K units or more)05:08
wolfsprauloh sure, but that's a completely different use case05:08
azonenbergBut as i said its the one i'm familiar with lol05:08
azonenbergGiven a wafer, go test it05:08
wolfspraulanyway I am just thinking practical how we go from your wafer to being able to integrate this into a bigger circuit05:08
wolfspraulnormally I think the packaging services are also small service providers05:08
azonenbergDepends a lot on how much ends up being doable05:09
wolfspraulyou send them a wafer, tell them which package you want, and they send you the packages back05:09
azonenbergIn the short term i am not expecting to make anything more complex than 4000 series05:09
azonenbergone die at a time05:09
johndmcmasteron that note, does anyone have a good reference on ohmic contacts05:09
johndmcmastercomposition materials and such05:09
azonenbergjohndmcmaster: That was on my list of things to research as well05:10
azonenbergi've heard something about platinum silicide being common05:10
azonenbergBut for the comb drive i'll be doing evaporated or sputtered metal over thermal oxide05:10
azonenbergiow the Si will be just a substrate and not passing any current at all05:10
wolfspraulso wait, you lost me05:10
wolfspraulwe will find some way to package it, let's move that aside now05:11
wolfspraulbut how about programmability05:11
wolfspraulwhat feature can come out of a 'comb drive'?05:11
azonenbergi was pushing that off until i got to CMOS05:11
azonenbergwolfspraul: On and off05:11
azonenbergIts meant to be used as part of a MEMS system05:11
azonenbergmoving around micromirrors etc05:11
azonenbergBy itself it's useless05:11
azonenbergAn x/y micromirror system would be next on the list of more complex devices to build05:12
wolfspraulimaging sensor?05:12
azonenbergno, it'd be a steerable mirror05:12
azonenbergA chunk of polished Si that can be tilted in various directions05:12
azonenbergBasically a single DLP pixel05:12
azonenbergWhich could also be used as the main element in a laser projector05:12
wolfspraulyes :-)05:12
azonenbergIt'd probably be sputtered in aluminum or something to make it nice and reflective (like telescope mirrors)05:13
azonenbergThats next on my todo list once i get a single comb drive built05:13
azonenbergSomething of that magnitude would actually be useful05:13
wolfspraulwell then, good luck!05:13
azonenbergand probalby not that difficult05:13
azonenbergCMOS is a lot harder due to the sensitivity to trace contaminants05:13
azonenbergThe micromirror would basically be two comb drives connected to a block of Si05:13
azonenbergthe only hard part would be figuring out how to make it tilt05:14
wolfspraulyes, anything usable first05:14
wolfspraulthat'd be great05:14
wolfspraulI will seriously try to make this part of a hacker board or product, in whatever stretch imagination that needs05:15
azonenbergAny kind of laser projector could use something like this05:15
wolfspraulwe take the fabrication of your dies to China! even if made one by one :-)05:15
azonenbergLol :P05:15
wolfspraul(just kidding)05:15
azonenbergMaybe to improve yields have two independent 1-axis tilt systems05:15
azonenbergthat would mean i don tneed to get two working drives on the same die05:15
azonenbergmake a working x axis, then a working y axis05:16
azonenbergand mount them 90 deg apart05:16
wolfspraulhow is this related to a galvanometer?05:16
azonenbergThe intended use case would be very similar05:16
azonenbergA DLP projector is basically a grid of tiny mirror galvos05:16
wolfspraulthere's a guy marcan in the #qi-hardware channel now working on his OpenLase project05:16
azonenbergthat PWM the reflected signal between "on" (poitned at the screen" and "off" (pointed off to the side)05:17
wolfspraulhe takes a normal 2 USD laser pointer, plus galvos and some software05:17
azonenbergAnd yeah, i saw05:17
azonenbergYou could quite possible use a MEMS galvo to do steering of the beam05:17
wolfspraulvery cool05:17
azonenbergthats something i'd say is realistic in the 1-2 year time frame05:17
azonenbergbased on how fast things are going now05:17
azonenbergOf course, production would be very limited lol05:17
wolfspraulin that time frame I can work on getting my manufacturing costs further down and efficiency further up05:18
wolfspraulnah don't worry. I will take this to China, then we optimize :-)05:18
azonenbergLol good luck finding budget for ASIC fab05:18
wolfspraulif Chinese are good at anything, it's process innovation05:18
azonenbergi know guys doing that and it isnt cheap05:18
azonenbergBut its an open project so if you do figure out a way to do something cool with my designs, more power to you05:18
wolfspraulwhy not make them in your very own process?05:18
azonenbergI could, it just couldnt be done en masse05:19
wolfspraulok one by one05:19
azonenbergRight now the biggest issue preventing more rapid production is my exposure system05:19
wolfspraulI would start with 1, then optimize05:19
azonenbergall of the chemical processing and deposition is parallel up to wafer level05:19
azonenbergBut i cant expose more than one die (or even part of a die depending on magnification) at a time05:19
wolfspraulunderstood. nothing better than a few nice challenges, right?05:20
azonenbergi need to work out a laser direct-write system or similar that will let me make a contact mask for a bunch of dies and expose them all at once05:20
azonenbergIts on the longer term todo list05:20
wolfspraulthe mems mirror may come in handy?05:20
azonenbergLol, maybe05:20
wolfsprauljust kidding, I think I understand it05:20
azonenbergi'm serious though05:20
wolfspraulwe will find ways to bring this out into real use cases though05:20
wolfspraulno matter how limited05:21
azonenbergWould be cool to show people my instruments and say that i am using a homemade chip into the tool05:21
azonenbergAnd sounds like a plan :)05:21
azonenbergMy next goal in the short term is to get 1/4 of a <110> 2-inch wafer prepped, spin coated in a very thin Ta2O5 layer (probably on the order of 50nm)05:21
azonenbergand coated with evaporated Cu05:22
azonenbergOn both sides05:22
azonenbergActually i might even do half a wafer05:22
azonenbergif the tantalum cooperates05:23
azonenbergMy proposed process would allow me to do all of the depositions at once and then etch top and bottom separately05:23
azonenbergThe one restriction is that once i open a hole in the hardmask i cant close it05:23
azonenbergin other words i'd need to do the bottom etch first and thin down to maybe 30 microns05:23
azonenbergthen open up the top (leaving bottom exposed) and etch both down and up05:23
azonenbergthe end result would be the fingers on top, parallel with the top surface, and about 15um thick05:24
azonenbergAt that point i'd strip the Ta2O5 layer, grow thermal oxide over the whole die (which i need to get a furnace for)05:24
azonenbergevaporate/sputter the whole die in metal and then do one last etch of metal 105:24
azonenbergat which point i'd have a completed, ready-to-test prototype05:25
wolfspraulI will follow...05:26
bart416azonenberg, what to do with a shitty old computer11:28
R0b0t1If you don't already know of a use for it I've found it is best thrown away.11:42
bart416Yeah but that costs money!11:45
bart416(Not a joke)11:45
berndjazonenberg, is it much harder to evaporate Cu than Al?  thinking of melting point vs vapour pressure here17:17
azonenbergberndj: Cu and Al can both be evaporated in the unit i've been using18:38
azonenbergI did Cu because it's more resistant to HF than Al18:38
azonenbergIn terms of a homebrew setup that's still a ways out18:40
bart416azonenberg, http://news.vanderbilt.edu/2011/08/nanodiamond/20:38
azonenberginteresting - CVD diamond for gates?20:42
azonenbergAnd lol, they require vacuum-sealed packaging to operate?20:42
azonenbergthose would be fun to reverse engineer20:42
azonenbergyou couldnt probe them without a vacuum chamber20:43
azonenbergwhich probably would mean a SEM too20:43
bart416You'd be using a SEM to reverse engineer them anyway :P20:44
azonenbergNot for larger process size devices20:44
azonenbergyou can do 350nm with an optical microscope if you're patient20:44
azonenberg250 and below pretty much needs a SEM though20:45
bart416Eventhough we never get to do it, us is always told to go for the electron microscope when debugging or reverse engineering electronic devices20:49
bart416For the simple reason that you can study the device in operation20:49
azonenbergYou can do that with an optical scope too though20:50
azonenbergJust not for really small stuff20:50
azonenbergi have probes and micropositioners already20:50
bart416Can you "see" electricity with an optical scope? :P20:50
azonenbergand plan to try studying a 4000 series chip live in the near future20:50
bart416Didn't think so20:50
azonenbergBut i can drop a needle down on top metal and plug that into a logic analyzer20:50
azonenbergAnd for somethign simple like a 4017 that should give good results20:52
azonenberg    0.190] breadboard kernel: sys_alloc_init: node = a00006e8, node->this_size.size = 15321:01
azonenberg[    0.197] breadboard kernel: sys_alloc_init: node->next_free = 0000000021:01
azonenberg[    0.204] breadboard kernel: sys_alloc_init: buf->first_free = 0000000021:01
azonenberg[    0.210] breadboard kernel: sys_alloc_init: node = a00006e8, node->this_size.size = 2995921:01
azonenberg[    0.218] breadboard kernel: sys_alloc_init: node->next_free = 0000000021:01
azonenberg[    0.224] breadboard kernel: sys_alloc_init: buf->first_free = a00006e821:01
azonenbergwhoops wrong tab21:01
azonenbergThough if you guys want to help me debug a sys_malloc() segfault on an embedded MIPS feel free :P21:01
R0b0t1Psh, allocate on the stack.21:05
azonenbergR0b0t1: I think the segfault is actualyl stack corruption that overflowed onto the heap21:06
azonenbergi only have 32KB of ram so its not hard21:06
R0b0t1Yeah, seems like it could easily be the problem21:06
azonenbergBut its tricky and nondeterministic21:06
azonenbergdoesnt happen (or at least not in the same spot) if i disable context switching21:06
R0b0t1You have 32kb of RAM, and you have context switching...?21:08
azonenbergThe PIC32 series has up to 128KB but the chip on my dev board atm is 3221:09
azonenbergand yes, i have multithreading21:09
azonenbergThe kernel will eventually be ported to a MIPS softcore on an FPGA with a ton of DRAM21:09
azonenbergbut i want to keep it light enough that i can use it on pic32 still if necessary21:09
azonenberga thread environment block is only a couple hundred bytes, maybe less21:09
azonenberg32x 32-bit registers = 128 bytes plus a little metadata21:09
R0b0t1Oh I keep forgetting programs aren't in RAM21:09
azonenberg512KB of memory mapped flash21:10
azonenbergcode is all execute-in-place21:10
azonenbergRight now the kernel image, libc, and a bunch of usermod test applications are around 80KB21:15
azonenbergand i'm using under 10KB of ram21:16
azonenbergThat will go down as i optimize and trim stuff out of the core image, there's a lot of test code21:16
azonenbergWell, this is going to be fun - tonight i'm going to try some more silicon etching21:46
azonenbergusing my new slow Cu etch to pattern the Cu, then HF through that onto the Ta2O5 and KOH21:46
azonenbergthis is a SU-8 polymer comb drive, 300 microns high, 30 micron wide fingers23:19
azonenbergsputtered in gold23:19
azonenbergSomething like this or smaller should be very doable23:20
azonenbergthis is 30um and my process should be good to 20 at least23:20
--- Mon Aug 8 201100:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!