#homecmos IRC log for Tuesday, 2011-09-27

azonenberg_workNeed to do some more testing but i think i just got this hardmask process worked out02:19
azonenberg_workthe trick is to use a ton of hcl and very little h2o202:19
adcwhat was the pi u sent me02:20
azonenberg_workadc: Which pic?02:21
azonenberg_workThe projct website has several02:21
azonenberg_workAnd i was having trouble with unetched Cr in trenches02:22
azonenberg_workI finally figured out an etch process that can remove it02:22
azonenberg_workNow i just need to see if my photoresist will survive it02:22
CIA-67homecmos r139 | trunk/lithography-tests/labnotes/azonenberg_labnotes.txt | Today's lab notes03:30
azonenberg_workGetting there - some undercut, it will need tweaking03:31
azonenberg_workBut in half an hour i expect to have results from a deep etch03:31
azonenberg_workMight need a little more Cr and a little less Cu03:34
azonenberg_workBut it's the right mix03:34
azonenberg_worklooks like Cr etches ~5nm/min while Si etches 2600nm/min03:36
azonenberg_workSo that would mean my adhesion layer is >5nm thick, more like 5003:36
azonenberg_workI'll need around a 190 minute etch to go through wafer, which means about a micron of Cr - very feasible to do with an evaporator03:37
azonenberg_workThen coat with a thin Cu layer to protect against oxidation03:37
azonenberg_workand actually, more like 500nm of Cr03:37
azonenberg_worksince i am going both ways03:37
azonenberg_workBut i think this is the process i will end up using03:37
CIA-67homecmos r140 | trunk/lithography-tests/labnotes/azonenberg_labnotes.txt | The rest of today's lab notes04:54
azonenbergDie J3 @ 100x http://i.imgur.com/ttSY7.jpg05:09
azonenbergand 400x http://i.imgur.com/eQdeM.jpg05:09
azonenbergThere was some undercut of the mask in a few spots, and the Cr hardmask was too thin so it got eaten through05:09
azonenbergAnd the etch was far from aligned to the crystal axes05:09
azonenbergBut for a first attempt i dont think its bad at all05:10
azonenbergBut we're a way out from having a working CPU05:16
azonenbergI have done full 8-bit designs on FPGAs (and am working on a 32-bit now)05:17
azonenbergIn terms of home fab so far we've done basic 2D copper interconnect wiring and i just got the beginnings of a silicon etch process working05:17
azonenbergtransistors are still further down the roadmap05:17
flowrhow many Mhz can someone get up to, with a homebrew CPU of transistors?05:24
azonenbergflowr: First you have to have specs for the transistors (switching speeds etc)05:26
azonenbergThen you have to do the gate-level design and find critical paths05:26
azonenbergthen you can compute the maximum delay along that path and that gives you the max frequency05:26
flowrcritical paths = bottle neck?05:27
azonenbergflowr: yeah, pretty much05:28
azonenbergbiggest delay between a pair of flipflops05:28
azonenbergincluding gates and wiring05:28
azonenbergMy guess is, >10 MHz using discrete transistors is not happening05:29
azonenbergeven thats high05:29
flowrhave you done any homebrew computers this way?05:29
azonenbergif you integrate it you may have a chance05:29
azonenbergI've used FPGAs05:29
azonenbergand am now working from the other end up on home chip fab05:29
azonenbergBut we dont even have functional transistors yet05:29
flowrhome chip fab? You mean, making your own ICs?05:30
azonenbergflowr: Yeah, thats what the channel is about lol05:30
flowrbut I mean, small ones :P05:30
flowror is that what you mean? :O05:31
azonenbergflowr: Small as in low gate count? Yes05:31
azonenbergSmall as in 22nm? No, our chips are huge dimension wise :p05:31
flowrthat's what I thought :P05:31
azonenbergmost of my test dies are a few mm across and the components are in a ~0.5mm disk in the middle05:31
azonenbergBut i havent actually fabricated anything more complex than test patterns in copper wiring05:31
azonenbergand, most recently, the beginnings of a MEMS device05:32
azonenberghoping for a simple working MEMS unit by end of this calendar year and logic gates some time in 201205:32
flowrwhat does one have to do, to get past 20Mhz, if one is soldering one's own gates on a circuit board?05:34
flowrDo you need pretty small transistors to get past 20Mhz?05:35
azonenbergflowr: No, you need them close together05:35
azonenbergI mean, getting a single gate that fast is easy05:35
azonenberggetting a string of gates end to end that fast is harder05:35
flowris it because the copper wire between them?05:36
azonenbergflowr: Routing delay is significant, yes05:36
azonenbergespecially in FPGA designs due to the less efficient routing it can be >50% of your total path delay05:36
azonenbergIf you have fast gates and long wires (like a breadboarded design) it will be even worse05:36
azonenbergMaking a working cpu is entirely feasible05:36
azonenbergmaking it fast is hard05:36
azonenbergI havent really looked into what transistor speeds etc would be in a homebrew fab process yet05:38
azonenbergits so far out that i have more pressing things to worry about05:38
azonenbergLike fully etching all of my chromium layers http://colossus.cs.rpi.edu/~azonenberg/images/homecmos/2011-09-20/die_i4_009.jpg05:39
azonenbergthis is an electron microscope shot of one of my chips taken last week05:39
azonenberglight gray is copper, dark is silicon05:39
azonenbergmedium is chromium (which is under the copper and should be completely removed in the trenches)05:39
flowrdoes the chromium conduct?05:40
azonenbergSlightly, yes05:41
azonenbergBut i wasnt using it as a wire05:41
azonenbergThe chrome is there because copper won't stick to things well by itself05:41
azonenbergit sticks to Cr and Cr sticks to a lot05:41
azonenbergSo a sandwich of the two is common05:41
flowrhow did you build this?05:41
azonenbergI also was using the sandwich as an etch mask for a future processing step05:42
azonenbergCant talk too much, its 2 am here and i need sleep, but heres a brief overview05:42
flowrk, maybe you should sleep :P05:42
azonenbergstart with blank <110> silicon wafer, score and snap into ~3mm pieces05:42
azonenbergdeposit Cr + Cu layers in a vacuum evaporator on campus (building one of my own is on the to-do list but is a few months out)05:43
azonenberggo back home, spin coat in photoresist05:43
azonenbergexpose, develop, etch in HCl with a few drops of 3% H2O2 added05:43
azonenbergthe Cr residue in this pic was the result of using too much H2O205:43
azonenbergthe Cu etched fine but the Cr formed CrO2 which is difficult to remove05:43
azonenbergand protected the underlying Cr from further etching05:44
Action: azonenberg AFKs05:44
azonenberglab notes are in the google code repo in the channel topic05:44
azonenbergif you want all of the gory details05:44
azonenbergjust look at recent changes and open up azonenberg_labnotes.txrt05:45
B0101hi Azonenberg12:21
azonenberghi B010112:22
azonenbergDid you see my pics from last night?12:22
azonenberg100 and 400x12:23
azonenbergthe mask was somewhat undercut so the lines got attacked12:23
azonenbergand the edges are rough since i didnt properly align everything to the crystal planes12:23
azonenbergB0101: where'd you go?12:35
azonenberg(08:23:19) azonenberg: http://i.imgur.com/ttSY7.jpg12:36
azonenberg(08:23:22) azonenberg: http://i.imgur.com/eQdeM.jpg12:36
azonenberg(08:23:25) azonenberg: 100 and 400x12:36
azonenberg(08:23:36) azonenberg: the mask was somewhat undercut so the lines got attacked12:36
azonenberg(08:23:59) azonenberg: and the edges are rough since i didnt properly align everything to the crystal planes12:36
B0101azonenberg: my computer crashed12:36
B0101this system is old\12:36
azonenbergBut as a first try i think its pretty good - i need a thicker Cr layer and possibly a shorter HCl etch before the KOH12:36
azonenbergthis was my first attempt at a deep KOH etch and it was at least somewhat successful :)12:36
B0101graet! :) btw, i have IRC logging on, so i can still see you links12:37
azonenbergWasnt sure if they had actually gone through lol12:37
azonenbergoften when a client crashes it freezes for a while and stops getting data12:37
azonenbergthen the OS closes the connection12:37
B0101but I got the link 3 seconds before it crashed so it would have been logged in12:38
azonenbergoh i see12:38
azonenbergAny luck with your squids?12:39
azonenberghow far have you gotten12:39
B0101i managed to fabricate a simple one12:40
azonenbergOut of what? And how small?12:40
azonenbergAnd have you tested it?12:40
CIA-67homecmos r141 | wiki/images/ (S7301905_fullres.jpg S7301905_thumb.jpg) | Uploaded two more images to wiki12:41
B0101Silicon, on a piece of silicon 5 cm12:41
B0101i tested it, but it stopped working an hour ago12:41
azonenbergSo it worked for a while, then stopped?12:41
azonenbergany idea why? If it was transistor based my first guess would be trace metal contamination12:42
B0101No ideas yet but I am investigating12:42
azonenbergwell make sure to post lab notes at some point so we can see12:42
azonenbergRemember, we dont just want a working process12:42
azonenbergwe want to know what went wrong and how you dealt with it12:43
azonenbergit could be very helpful to someone else having a similar problem12:43
azonenbergsimilar symptoms might indicate a similar problem, etc12:43
B0101The only thing I know is that it changed from rainbow color to silver and thats all i know12:43
azonenbergHmm, rainbow color indicates a thin transparent layer (oxide etc)12:44
azonenbergsilver would mean either much thicker or none12:44
azonenbergdid it get hot in operation at all?12:44
B0101I don't know if it is the voltage12:46
B0101I applied 5V AC to it12:46
azonenbergNo idea what they expect12:47
azonenbergthis is your research lol12:47
B0101I'm gonna go back to the lab to investigate further12:47
azonenbergquantum computing is not my area of interest atm12:47
azonenbergAnd ok12:47
B0101See you in 18+ hours12:47
--- Wed Sep 28 201100:00

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