| diginet | sivoais: what are you doing here? | 01:17 |
|---|---|---|
| sivoais | haha, hey diginet! I'm an engineer. I like engineering things. | 02:00 |
| diginet | I'm not an engineer (lol), but I've talked with azonenberg a little about homecmos | 02:38 |
| diginet | I think having a DIYable chip fab would be just about the greatest thing ever | 02:39 |
| diginet | we were talking earlier actually (in PM), and I thought of something that seemed odd and he didn't know either, maybe you can answer | 02:40 |
| diginet | sivoais: why is tungsten (often) used for vias in CMOS? I know that copper has issues with electromigration, but why not aluminum? | 02:43 |
| sivoais | diginet: no idea. I'm not a materials person. Would be a good thing to look through the literature for | 02:46 |
| diginet | good idea | 02:49 |
| Action: diginet googles | 02:49 | |
| Sync_ | diginet: I guess it has to withstand KOH etches and be easily depositable | 05:37 |
| diginet | Sync_: yeah, I looked up some papers and they basically said something similar | 05:38 |
| Sync_ | but it seems cu is possible after all | 05:39 |
| diginet | yeah | 05:39 |
| Action: Sync_ is going to kill his prof.... | 05:40 | |
| azonenberg | Sync_: Nobody uses KOH for CMOS | 05:40 |
| azonenberg | TMAH maybe | 05:40 |
| azonenberg | but Cu survives exposure to low-concentration TMAH used as developer | 05:40 |
| azonenberg | Al is slightly attacked | 05:40 |
| diginet | you know, I was thinking, wasn't the whole point of silicon that it has a (relatively) good native oxide? If high-k dielectrics are replacing SiO2, why even both anymore | 05:43 |
| diginet | the problem is finding an alternative that isn't made of unobtainium or something stupid like Arsenic | 05:43 |
| Sync_ | I thought KOH was still used for bulk etching | 05:45 |
| Sync_ | diginet: germanium is the answer | 05:45 |
| diginet | really? why do you think so | 05:46 |
| Sync_ | becaue I work in it | 05:46 |
| Sync_ | we're doing Ge on Si | 05:46 |
| Sync_ | and GeSi stuff | 05:46 |
| Sync_ | which is easily integratable in current cmos processes | 05:47 |
| diginet | really | 05:47 |
| Sync_ | your phone does it | 05:47 |
| diginet | I wasn't questioning you, I was just curious | 05:47 |
| diginet | I don't have a phone :P | 05:48 |
| diginet | I knew it was used for RF stuff, but could it replace silicon in general? | 05:48 |
| Sync_ | well, that's the hope | 05:48 |
| Sync_ | because slowly making ever so slightly smaller devices is getting interesting | 05:49 |
| diginet | but isn't germanium a lot more fragile than silicon? | 05:50 |
| Sync_ | to a degree yes but that's why the GeSi stuff is so interesting | 05:50 |
| Sync_ | but really, the person who finds a way to integrate optoelectronics in cmos will take the deal | 05:51 |
| Sync_ | that'd solve a lot of interconnect issues | 05:51 |
| diginet | optical innerconnects? | 05:51 |
| Sync_ | yes | 05:52 |
| diginet | as I understand it, the main issue right now is parasitic capacitance, no? | 05:52 |
| Sync_ | yes | 05:52 |
| diginet | what about GaAs? | 05:53 |
| Sync_ | that's not really compatible with cmos | 05:53 |
| diginet | ah | 05:53 |
| diginet | isn't germanium like insanely expensive though? | 05:53 |
| Sync_ | that's why GeSi + C HEMTs are so cool for HF | 05:53 |
| diginet | C, like graphene? | 05:54 |
| Sync_ | because you can put them on die with the rest of the chip | 05:54 |
| Sync_ | no just plain carbon | 05:54 |
| diginet | HEMT are the one's which use 2d electron gas right? | 05:54 |
| Sync_ | yeah germanium is spendy, that's why you take a Si wafer and deposit stress free germanium | 05:54 |
| diginet | what is their main usage? in theory could you use them to speed up a regular chip like a CPU? | 05:55 |
| Sync_ | mobile phones, RF stuff | 05:55 |
| Sync_ | they have transition frequencies over 500GHz now | 05:55 |
| diginet | okay this is weird, according to wikipedia, SiGe has lower junction leakage because it has a /lower/ band gap | 05:57 |
| diginet | I thought it was a /higher/ bandgap which would lead to lower leakage? | 05:57 |
| Sync_ | yes | 05:57 |
| Sync_ | magic of sige | 05:57 |
| Sync_ | no, really it does some really strange things | 05:57 |
| diginet | wait, how does that even work though? | 05:58 |
| Sync_ | in hemts you can do strange stuff like vary the ge concentration through the channel | 05:58 |
| diginet | if the energy required to jump from the valence band to the conduction band is lower, shouldn't that mean an electron is more likely to make the jump? | 05:59 |
| diginet | why are HEMTs not used for non-RF stuff then? | 05:59 |
| Sync_ | price and that cpu speed is as you said limited by the intereconnects | 05:59 |
| diginet | but they are otherwise superior in nearly every way? | 06:00 |
| Sync_ | yup | 06:05 |
| Sync_ | but the main thing about them is that you can integrate them cheaply in a cmos process | 06:05 |
| diginet | what makes them more expensive than regular FETs? | 06:07 |
| Sync_ | well | 06:08 |
| Sync_ | putting ge in? | 06:08 |
| Sync_ | :D | 06:08 |
| Sync_ | sorry, exam in 1h by brain is a bit wonky | 06:09 |
| diginet | no problem :P | 06:09 |
| berndj | diginet: electromigration goes with atomic mass. aluminium is worse than copper because it's lighter. tungsten is better than both. maybe you need high current densities at contacts, so you use tungsten? | 20:47 |
| diginet | Al is worse than Cu? Huh. I didn't know that | 20:48 |
| berndj | [citation needed] and all that. don't take my word for it | 20:50 |
| diginet | gotcha | 20:51 |
| berndj | but yeah, as far as i understood it people moved from Al to Cu because Cu could take higher current density, which mattered as feature sizes got smaller | 20:58 |
| diginet | yeah | 21:06 |
| diginet | does anyone here know a good place to buy tantalum pentoxide? | 22:17 |
| diginet | does not need to be high purity, ~97-98% is fine | 22:18 |
| berndj | azonenberg, was using it some time back. if you're lucky the channel logs might find it for you? | 22:28 |
| azonenberg | diginet: You can get a solution of it from Emulsitone | 22:49 |
| azonenberg | well | 22:49 |
| azonenberg | tantalum chloride in ethanol | 22:49 |
| azonenberg | spin coat and bake and you get a film of Ta2O5 | 22:49 |
| azonenberg | It cost me $500 for 4 ounces | 22:49 |
| diginet | yikes | 22:49 |
| azonenberg | I wanted to use it as a MEMS hardmask | 22:49 |
| azonenberg | I might still try that once i solve my photoresist adhesion issues | 22:49 |
| azonenberg | HMDS might help with that | 22:49 |
| diginet | surely there have to be cheaper sources of Ta2O5 | 22:51 |
| azonenberg | In what form | 22:52 |
| azonenberg | In my issue the issue was the minimum order | 22:52 |
| azonenberg | i only needed a few drops | 22:52 |
| diginet | powder | 22:53 |
| azonenberg | they wouldnt sell me less than 4oz of the 16% solution | 22:53 |
| azonenberg | What are you trying t odo with it? | 22:53 |
| azonenberg | High-K? | 22:53 |
| azonenberg | hardmask? | 22:53 |
| azonenberg | optical coating? | 22:53 |
| diginet | no, I decided to use it for my WORM storage thing | 22:53 |
| azonenberg | Oh | 22:53 |
| azonenberg | Hmm | 22:53 |
| azonenberg | What advantages does it have over SiC | 22:53 |
| diginet | I would buy the metal, but it's fairly easy to refine, and supposedly a lot cheaper | 22:53 |
| diginet | no, Ta for the metal | 22:53 |
| azonenberg | Oh | 22:53 |
| azonenberg | So you really want metallic Ta | 22:53 |
| diginet | yeah | 22:54 |
| azonenberg | How much, what purity? | 22:54 |
| diginet | but metallic Ta costs more, so I figured I'd save some money and DIY the refining, I've been meaning to make an FFC reactor anyway | 22:54 |
| diginet | not high, 95+ would work | 22:54 |
| azonenberg | sec | 22:54 |
| diginet | maybe 10g? (10g metal, not sure how much oxide I'd need for that) | 22:54 |
| diginet | tantalum has such good properties, probably the best for my purposes of any metal that isn't some insane 10-component superalloy | 22:55 |
| azonenberg | Which properties | 22:56 |
| diginet | corrosion resistance and high melting point namely | 22:56 |
| diginet | tungsten melts at a high temperature as well, but it oxidizes at above 300 C | 22:56 |
| azonenberg | How about Au or Pt or Pd? | 22:56 |
| azonenberg | Ignoring cost, just thinking mechanical properties | 22:57 |
| diginet | more expensive | 22:57 |
| azonenberg | or Nb | 22:57 |
| diginet | and Ta is better than all of those | 22:57 |
| diginet | hmm | 22:57 |
| diginet | Nb is an option | 22:57 |
| azonenberg | Looking up corrosion resistance | 22:57 |
| diginet | DON'T YOU MEAN COLUMBIUM YOU COMMIE | 22:57 |
| diginet | /troll | 22:57 |
| azonenberg | its very similar to Ta, wiki says it melts at 2477C | 22:57 |
| azonenberg | Slightly less corrosion resistant | 22:58 |
| azonenberg | but... | 22:58 |
| azonenberg | http://www.tedpella.com/vacuum_html/vacwire.htm | 22:58 |
| azonenberg | 10 grams of Nb in 1/4" pellets, 99.95% purity | 22:58 |
| azonenberg | $28.50 | 22:58 |
| diginet | yeah | 22:58 |
| diginet | that's kind of hard to beat | 22:59 |
| azonenberg | http://www.elementsales.com/ecoins_desc.htm#Ta | 22:59 |
| azonenberg | $16 for 7.2 grams | 22:59 |
| azonenberg | 99.9% | 22:59 |
| diginet | hmm | 23:00 |
| diginet | http://www.engineeringtoolbox.com/metal-corrosion-resistance-d_491.html | 23:01 |
| diginet | look at hastelloy C | 23:01 |
| diginet | insane | 23:01 |
| azonenberg | beer, lol | 23:01 |
| azonenberg | can you imagine being in a lab evaluating the corrosion resistance of materials to beer? | 23:02 |
| azonenberg | hydrofluoric acid... | 23:02 |
| azonenberg | ratred 3 against wet chlorine gas | 23:02 |
| diginet | hastelloy C? | 23:02 |
| azonenberg | You need to be able to do fine patterning of it too though | 23:02 |
| azonenberg | Ta can be patterned precisely | 23:02 |
| azonenberg | sorry, i meant 2 | 23:02 |
| diginet | yeah | 23:02 |
| diginet | ah | 23:03 |
| diginet | what about Nb? | 23:03 |
| azonenberg | Alloys may not be the best bet | 23:03 |
| azonenberg | I havent looked at Nb but Ta is used in IC fabrication as a barrier/adhesion layer for Cu | 23:03 |
| azonenberg | So you can get deep sub-nanometer patterning and fine grain sizes | 23:03 |
| diginet | hmm | 23:04 |
| diginet | apparently alloying Nb with Zr and W can substantially improve it's qualities | 23:04 |
| azonenberg | Bear in mind you need to deposit the alloy as a thin film somehow, right? | 23:05 |
| azonenberg | Not all deposition processes are amenable to this | 23:05 |
| azonenberg | evaporation for example does not work well with alloys | 23:05 |
| azonenberg | i think sputtering does | 23:05 |
| diginet | it's not thing film though | 23:06 |
| azonenberg | isnt it a thin film on a glass substrate? | 23:06 |
| diginet | that wasn't my intent | 23:06 |
| diginet | I was going to cast an approx 5mm thick plate | 23:06 |
| azonenberg | And read back how? | 23:07 |
| azonenberg | the benefit of metal film on a clear substrate is easy optical readout | 23:07 |
| diginet | with a microscope | 23:08 |
| diginet | this isn't a digital medium | 23:08 |
| azonenberg | I'm just saying, tiny variations in surface topography | 23:08 |
| azonenberg | are less good contast than presence or absence of an opaque material on a clear substrate | 23:08 |
| azonenberg | contrast* | 23:08 |
| diginet | http://i.imgur.com/Fioe5Rd.png | 23:09 |
| azonenberg | Yea, but I pictured your design as being a film a few nm thick | 23:09 |
| azonenberg | of metal | 23:09 |
| azonenberg | maybe a micron at most | 23:09 |
| diginet | the plate is cast seperately | 23:09 |
| azonenberg | then a thick substrate and a slightly thinner cap | 23:09 |
| azonenberg | why? | 23:09 |
| diginet | because it gives me more options on the composition | 23:10 |
| azonenberg | I'm just thinking in the end you have to be able to read it | 23:10 |
| diginet | you can, the HD rosetta process, while it is a film on glass, is all nickel | 23:10 |
| diginet | i.e. you can't see through the Ni film | 23:11 |
| azonenberg | Oh | 23:11 |
| azonenberg | I was envisioning something similar to standard chrome-on-glass semiconductor masks | 23:11 |
| azonenberg | except with a second layer on the top surface for protection | 23:11 |
| diginet | http://www.norsam.com/rosetta.html | 23:12 |
| diginet | see | 23:12 |
| diginet | haha, no, lol | 23:12 |
| azonenberg | Such that you could backlight it and read out optically | 23:12 |
| azonenberg | with extremely high contrast | 23:12 |
| diginet | remember, I'm planning on making it holographic though | 23:12 |
| azonenberg | Hmm | 23:13 |
| diginet | the read out will be with a laser | 23:13 |
| diginet | I won't be directly writing the pixels to the plate, rather I will computer the diffraction gratings and then draw those out | 23:13 |
| azonenberg | oh, i see | 23:13 |
| azonenberg | I was talking about if you wanted direct optical imaging | 23:13 |
| diginet | yeah | 23:14 |
| azonenberg | of either binary data or text or graphics | 23:14 |
| azonenberg | something like that could be read with nothing but a microscope | 23:14 |
| diginet | yeah, but I think a non-holographic version of mine could as well | 23:14 |
| diginet | I could fill in the troughs with something else | 23:14 |
| diginet | to increase contrast | 23:14 |
| azonenberg | The ideal structure for my idea would be a clear SiC wafer or similar with a very thin patterned metal film | 23:15 |
| azonenberg | then more | 23:15 |
| azonenberg | Ta2O5 might actually be a good surface coating lol | 23:15 |
| azonenberg | Almost nothing but HF will corrode it | 23:15 |
| diginet | indeed | 23:15 |
| azonenberg | and even HF does so very slowly | 23:15 |
| diginet | that's why I decided against W, a lot of stuff corrodes W | 23:15 |
| azonenberg | http://orbit-dtu-dk.cvt.dk/services/downloadRegister/4645876/Christensen.pdf | 23:16 |
| azonenberg | weeks of immersion in hot KOH | 23:16 |
| azonenberg | etch rates were too slow to measure | 23:16 |
| diginet | wow | 23:17 |
| azonenberg | That was why i decided to use it as a hardmask for KOH etches | 23:17 |
| azonenberg | my problem was, it was TOO etch resistant | 23:17 |
| azonenberg | i couldn't pattern it | 23:17 |
| azonenberg | my photoresist would peel off the surface before it etched through | 23:17 |
| azonenberg | HMDS might help with that | 23:17 |
| azonenberg | on my "future research" list | 23:21 |
| azonenberg | too many other things going on lol | 23:21 |
| --- Sat Feb 16 2013 | 00:00 | |
Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!
