|
I dunno, talk about microporcessor poo poo here or something? E1: Previous HW thread: https://forums.somethingawful.com/showthread.php?threadid=3683831 E2: ate all the Oreos posted:it's simple, p-types say Please fill my holes, and n-types say Nice! and agree, and then when two of them love each other very much they meet up in a seedy hotel and have passionate sex and afterwards feel very depleted while smoking in bed and thinking about what they just did, but a shot of electricity (the MDMA of the atomic realm) can temporarily fix this depletion and get them raring to go again Schadenboner fucked around with this message at 03:18 on Jul 6, 2018 |
#
?
Jul 5, 2018 13:25
|
|
|
|
| # ? Apr 26, 2024 21:27 |
|
|
I too would love to hear about fabs. One question off the top of my head: how far does the cottage chip industry lag behind the state of the art? If I wanted to design a custom processor, who would I call to manufacture it, and what would it look like compared to these insane 14nm, 7nm etc stuff (Assuming anyone would bother? How much is the minimum order for this kind of thing?) The whole industry is interesting and basically a mystery to me, I'm so far from the metal I've forgotten what it even looks like.
|
|
#
?
Jul 5, 2018 13:31
|
|
|
Yeah, I was an Economics major* who got into CJing around campus to make beer money and somehow ended up going career IT so all this engineering poo poo is both confusing and interesting in roughly equal measure. Hence, the thread. *: Economics was in Liberal Arts rather than B-School** so I got to learn all sorts of "make interesting smalltalk at cocktail parties" poo poo, rather than, you know: skills? **: I would probably be a lot better-off if I had done MIS. 
Schadenboner fucked around with this message at 13:43 on Jul 5, 2018 |
|
#
?
Jul 5, 2018 13:37
|
|
|
ADINSX posted:I too would love to hear about fabs. One question off the top of my head: how far does the cottage chip industry lag behind the state of the art? If I wanted to design a custom processor, who would I call to manufacture it, and what would it look like compared to these insane 14nm, 7nm etc stuff (Assuming anyone would bother? How much is the minimum order for this kind of thing?) for the right price it's not too hard to do a generation behind whatever intels cutting edge is
|
|
#
?
Jul 5, 2018 13:39
|
|
|
you'd also have to unload a dump truck of cash for cadence
|
|
#
?
Jul 5, 2018 13:40
|
|
|
normal cottage industry asics run through MOSIS and I think the min is 40k unless you're a university then it's free I think?
|
|
#
?
Jul 5, 2018 13:41
|
|
|
but that's on garbage old processes like 250nm or something. great if you're making analog parts but lol if it's anything real digital
|
|
#
?
Jul 5, 2018 13:42
|
|
|
is this the thread where we get to lol about globalfoundries
|
|
#
?
Jul 5, 2018 13:57
|
|
|
dragon enthusiast posted:is this the thread where we get to lol about globalfoundries globalfoundries was my single largest software sale and then they refused to pay on the date as stipulated in the contract and terminated the project. i am thoroughly enjoying their raft of poo poo. hope these job cuts are primarily in IT.
|
|
#
?
Jul 5, 2018 14:39
|
|
|
ADINSX posted:I too would love to hear about fabs. One question off the top of my head: how far does the cottage chip industry lag behind the state of the art? If I wanted to design a custom processor, who would I call to manufacture it, and what would it look like compared to these insane 14nm, 7nm etc stuff (Assuming anyone would bother? How much is the minimum order for this kind of thing?) first you need to understand that the 14nm and such refers to the smallest possible feature that can be made. but even in high performance cutting edge designs that resolution is only used at a few key steps, the rest are run using older techniques because they are cheaper. a 14nm fab will also be making larger chips to soak up any capacity they have once they've maxed out capacity on their highest resolution systems. cheapest would be the fabs which specialize in power semiconductors, ruggedized, and radiation-hardened designs which have gigantic dimensions and can use the cheapest photolithography systems. minimum feature sizes would be around 350nm. think companies like ON Semiconductor (running old Motorola fabs). that said any fab is going to have a pretty high and expensive minimum order due to the need to have photolithography masks made for every patterning layer. even the cheapest masks are several hundred dollars each and you'll need at the very very minimum (like single transistor with contacts) a half dozen, with complicated logic arrangements potentially pushing 40+.
|
|
#
?
Jul 5, 2018 15:14
|
|
|
there are prototyping companies where instead of having to do tooling for an entire wafer, they'll take a bunch of prototypes from different people and put together masks to make all of them in smaller quantities on a single wafer. this brings your up-front costs down from "ridiculous" to merely "exorbitant"
|
|
#
?
Jul 5, 2018 15:40
|
|
|
btw, custom chip development sucks and you shouldn't do it unless you absolutely have to
|
|
#
?
Jul 5, 2018 15:45
|
|
|
Bloody posted:btw, custom chip development sucks and you shouldn't do it unless you absolutely have to this times a million. i inherited a project years ago that the guy two predecessor pms back had decided to do a custom chip instead of just buying off the shelf and it was a disastrous shitpile the whole way through. finally i'd had enough and got my bossmasters to back me on scrapping the custom trash and just going with something from cypress that had been released in the meantime.
|
|
#
?
Jul 5, 2018 17:08
|
|
|
from someone who has no idea how this poo poo works, what design package do you use to develop custom chips? do the packages help you design by short-cutting calculations and applying physics models and stuff or is it straight Wild-West “draw it, slam it into the fab lol here comes my bits” stuff? also do people hide dicks, distended balls or goatse or something in chip designs for shits ‘n giggles? if not I will hate CS people more so than I do now
|
|
#
?
Jul 5, 2018 18:58
|
|
|
Midjack posted:this times a million. i inherited a project years ago that the guy two predecessor pms back had decided to do a custom chip instead of just buying off the shelf and it was a disastrous shitpile the whole way through. finally i'd had enough and got my bossmasters to back me on scrapping the custom trash and just going with something from cypress that had been released in the meantime. why you throw chip
|
|
#
?
Jul 5, 2018 19:02
|
|
|
Ripoff posted:
this happened all the time, though it's somewhat less common now with the proliferation of automated layout tools.
|
|
#
?
Jul 5, 2018 19:28
|
|
|
Ripoff posted:from someone who has no idea how this poo poo works, what design package do you use to develop custom chips? do the packages help you design by short-cutting calculations and applying physics models and stuff or is it straight Wild-West “draw it, slam it into the fab lol here comes my bits” stuff? they usually combine CAD software like cadence where you're drawing the patterns that actually go on the wafer, with circuit simulation software that has all the physical models and simulation like spice the basic steps are DRC (design rule check) - this looks at your CAD drawing and looks for violations like "yo these copper wires can't be this close to one another" or "you idiot you made a transistor shape that won't actually work". the rules are all written in a variant of lisp lol LVS (layout versus schematic) - this turns the CAD layout and the circuit schematic into the equivalent of electrical bytecode and checks to see if they're actually the same there's some other steps like parasitic extraction and design for manufacturing that help with design fabs will release process design kits (PDKs) to circuit designers that have the CAD libraries and circuit elements built in
|
|
#
?
Jul 5, 2018 19:44
|
|
|
idk if its just me but 'HOTCHIPS' is a really good name for a conference https://www.hotchips.org
|
|
#
?
Jul 5, 2018 22:19
|
|
|
ate all the Oreos posted:it's simple, p-types say Please fill my holes, and n-types say Nice! and agree, and then when two of them love each other very much they meet up in a seedy hotel and have passionate sex and afterwards feel very depleted while smoking in bed and thinking about what they just did, but a shot of electricity (the MDMA of the atomic realm) can temporarily fix this depletion and get them raring to go again
|
|
#
?
Jul 6, 2018 03:10
|
|
|
fritz posted:idk if its just me but 'HOTCHIPS' is a really good name for a conference https://www.hotchips.org
|
|
#
?
Jul 8, 2018 07:41
|
|
|
how come everyones planning to go from 193 nm photolithography to 13.5 given how hard it is? why not something in between also what about non silicon semiconductors for c/gpus?
|
|
#
?
Jul 8, 2018 13:38
|
|
|
What's the best chipset these days? I'm thinking about building a VR capable machine because my life is dull and meaningless and I need a distraction.
|
|
#
?
Jul 8, 2018 14:11
|
|
|
PyPy posted:What's the best chipset these days? I'm thinking about building a VR capable machine because my life is dull and meaningless and I need a distraction. the problem is GPU prices are all hosed up cause cryptocurrency mining
|
|
#
?
Jul 8, 2018 14:45
|
|
|
ConanTheLibrarian posted:also what about non silicon semiconductors for c/gpus? I'm not really in the know, but I'd be shocked if they actually incorporated non-silicon semiconductor channel materials into the computer chip manufacturing processes. I have heard that the usual candidate semi-conductors for doing this, indium gallium arsenide & germanium, while they could work better than silicon at longer gate lengths, they wouldn't work as well at really really short gate lengths. The same physical characteristic which was usually thought of as a boon at longer channel lengths might become a liability at really really short channel lengths. A hot area of research is on these new-to-electronics semiconductor materials which are naturally thin ('transition metal chalcogenides'). The idea being that if these materials are naturally thin, they would exhibit less variation than normal semiconductors when made into very small transistors. I'm not up to date, but when I checked, these materials aren't really performing as well as silicon in terms of on-current density and major manufacturing issues would have to be solved before they could get incorporated with computer chip technology. The way researchers currently make transistors from these 'transition metal chalcogenide' materials is 100% at odds with the idea of integrated circuitry and mass production--currently they place the non-silicon transistor material onto silicon wafers/dies in a really uncontrolled manner, and the material gets placed onto the silicon wafer in random locations, and so they have to make transistors out of the material in random locations. They don't have processes for depositing controlled films of these materials. Also, there already is an entire non-silicon III-V semiconductor industry which is mostly oriented towards supplying chips for special high-speed analog functions in electronic test equipment, wireless communication, and radar systems. If you look at what they have currently and try to see how well they would fare for computer chips, you'd see that they wouldn't do very well. The non-silicon semiconductor technology is relatively speaking, stuck in the Stone Age and is more oriented towards special high-speed analog functions, where the technical requirements are different from computer chips. silence_kit fucked around with this message at 15:34 on Jul 8, 2018 |
|
#
?
Jul 8, 2018 15:30
|
|
|
right now the leading edge for SiGe BiCMOS is 180nm, this is the kind of stuff that goes into front end modules and power amplifiers on your cell phone
|
|
#
?
Jul 8, 2018 16:38
|
|
|
CRIP EATIN BREAD posted:the problem is GPU prices are all hosed up cause cryptocurrency mining
|
|
#
?
Jul 8, 2018 16:48
|
|
|
fun fact for those of us who still had old school assumptions about chips: small nanometer silicon chips degrade with use. your CPUs have a limited lifespan before they start developing errors in their caches and other failures. that lifespan is coming down to a small number of years.
|
|
#
?
Jul 8, 2018 18:14
|
|
|
The Management posted:fun fact for those of us who still had old school assumptions about chips: small nanometer silicon chips degrade with use. your CPUs have a limited lifespan before they start developing errors in their caches and other failures. that lifespan is coming down to a small number of years. use the turbo button judiciously.
|
|
#
?
Jul 8, 2018 18:21
|
|
|
The Management posted:fun fact for those of us who still had old school assumptions about chips: small nanometer silicon chips degrade with use. your CPUs have a limited lifespan before they start developing errors in their caches and other failures. that lifespan is coming down to a small number of years. Are you talking about 'silicon aging'? Do you know what causes this? Is it https://en.wikipedia.org/wiki/Negative-bias_temperature_instability , where the transistor threshold voltage shifts over time? Or are you talking about something else?
|
|
#
?
Jul 8, 2018 18:28
|
|
|
silence_kit posted:Are you talking about 'silicon aging'? Do you know what causes this? Is it https://en.wikipedia.org/wiki/Negative-bias_temperature_instability , where the transistor threshold voltage shifts over time? Or are you talking about something else? I’m not a silicon expert to know why it happens, I just have to deal with the fallout. overdriven voltage, high frequency, high temperatures plus a few years and you see some interesting failures.
|
|
#
?
Jul 8, 2018 18:43
|
|
|
The Management posted:I’m not a silicon expert yeah. same. real bewbs ONLY.
|
|
#
?
Jul 8, 2018 18:47
|
|
|
silence_kit posted:Are you talking about 'silicon aging'? Do you know what causes this? Is it https://en.wikipedia.org/wiki/Negative-bias_temperature_instability , where the transistor threshold voltage shifts over time? Or are you talking about something else? I would expect it is mostly due to dopant and defect migration. it doesn't take as long for the atoms doping your gate area to disperse or for defects at grain boundaries in a metal conductors to agglomerate together via quantum and other effects when your transistor is smaller and running hotter. to make it worse one of the bad things about finfets and silicon in insulator designs is the transistor parts are not directly connected to the crystalline silicon substrate. the heat has to go through a bunch of other structures which aren't as thermally conductive and that raises the temperature of the transistor junctions themselves. that also makes it a lot more difficult to measure the temperature of the finfets. you used to be able to just form a pair of diodes in the middle of your transistor array to make a bandgap temperature sensor. since the sensor was well coupled with the transistors due to the highly thermally conductive single crystal silicon the measured temperature was quite close to the actual transistor temperature. that doesn't work as well now.
|
|
#
?
Jul 8, 2018 18:59
|
|
|
Shifty Pony posted:I would expect it is mostly due to dopant and defect migration. it doesn't take as long for the atoms doping your gate area to disperse or for defects at grain boundaries in a metal conductors to agglomerate together via quantum and other effects when your transistor is smaller and running hotter. This sounds like startrek technobabble filler, next picard will ask "what can we do to lengthen chip life" and geordi will think for a second and have a solution. Picard will say "make it so"
|
|
#
?
Jul 8, 2018 19:21
|
|
|
atoms jiggle when hot. sometimes they jiggle just right to trade places with the atom next to them. when your transistor is half the size of what it used to be the atoms that make it work only have to trade places half as far before they are someplace where they are going to gently caress things up. oh and wrapping the entire thing in insulation makes it worse because the atoms jiggle harder. thanks to probability fuckery the actual increase in likelihood of it happening is some crazy equation with a lot of exclamation marks in it that I don't feel like calculating out but let's just say that as you shrink things the effect isn't noticeable until it very suddenly is a huge problem.
|
|
#
?
Jul 8, 2018 19:47
|
|
|
|
|
#
?
Jul 8, 2018 19:51
|
|
|
ADINSX posted:This sounds like startrek technobabble filler, next picard will ask "what can we do to lengthen chip life" and geordi will think for a second and have a solution. Picard will say "make it so" lol, just lol, if u dont have a laymans level understanding in solid state physics
|
|
#
?
Jul 8, 2018 20:02
|
|
|
please stop reducing the lifespan of my chips
|
|
#
?
Jul 8, 2018 20:06
|
|
|
ADINSX posted:This sounds like startrek technobabble filler, next picard will ask "what can we do to lengthen chip life" and geordi will think for a second and have a solution. Picard will say "make it so" Haha when TNG came out, I think it was a pretty good time for the semiconductor industry I haven't seen the new Star Trek show, but I suspect the techno-babble in that one would definitely have a different flavor and they probably wouldn't be talking about transistors Shifty Pony posted:I would expect it is mostly due to dopant and defect migration. it doesn't take as long for the atoms doping your gate area to disperse or for defects at grain boundaries in a metal conductors to agglomerate together via quantum and other effects when your transistor is smaller and running hotter. Ok, this makes sense to me, at least. Stress creates transistor threshold voltage shifts over time (you attribute it to dopant migration out of/into the transistor channel, the wikipedia article I linked said that people used to attribute it to the hydrogen passivation of the interfacial silicon dangling bonds wearing off), which creates errors in circuits unless you throw away circuit performance/power consumption or increase cost to be more tolerant of the threshold voltage shifts. The threshold voltage shifts are more noticeable in more modern technologies due to the lower supply voltages, and might be exacerbated by the smaller channel areas and hotter channel temperatures due to the transistors tending to be more thermally isolated in the modern fin-based and SOI designs. silence_kit fucked around with this message at 20:10 on Jul 8, 2018 |
|
#
?
Jul 8, 2018 20:07
|
|
|
Midjack posted:use the turbo button judiciously. threads young, but this is a good title e: I have this TI 75 year book. It doesn't have anything too interesting, as it's corporate propaganda, but I skimmed and took some pictures with my oldphone for the first page. Cover  Inside cover.    Random images     Buddy, they won't even let me be the Chapstik. Single page blurb on the GOAT. Classic cool.PyPy fucked around with this message at 22:45 on Jul 8, 2018 |
|
#
?
Jul 8, 2018 21:50
|
|
|
|
| # ? Apr 26, 2024 21:27 |
|
|
PyPy posted:
In it voted 5
|
|
#
?
Jul 9, 2018 15:27
|
|