Semiconductors are more than just processors and GPUs(blog.robertelder.org) |
Semiconductors are more than just processors and GPUs(blog.robertelder.org) |
They are all GONE. No stock of anything (except the crappy ones, super-tiny packages, high Vgs(th) or high Rds(on) and other leftovers).
I've never seen anything like this, it's kind of frightening. Like walking into a grocery store and seeing the aisles all EMPTY except for a few scraps.
I don't even know where they all went. It's not like you need a TSMC slot to make a FET.
And whatever you look up, Chinese brokers have 10K-50K pieces of them for $25 each. Don't know what to think of that, either.
1. Our contract manufacturer calls in a panic no longer able to obtain/was shorted on a shipment of part XYZ. XYZ is increasingly becoming random "jellybean" parts like MOSFETS, oscillators, to slightly-more-complicated but not "fancy" stuff like serial transceivers, USB stuff, NOR flash, load switches. TI is the bane of my existence currently.
2. Search for a drop-in or near drop-in replacement. There are none, because that's what everyone's doing.
3. Search for alternative designs. Maybe the component is in distributor's stock (Digikey, Mouser, Newark, etc), maybe it's not.
4. Test the alternative design. By the time I receive parts, prototype, test, guess what? Can't get those parts anymore. Go back to step #2.
5. Fall behind on all of my other NPD responsibilities. Stress, burnout, acceptance. Lament not going into another engineering field. Feel bad about my midwest metro area compensation in comparison to a bunch of Silicon Valley SWEs on website.
6. GOTO #1
On new designs, I find a part in stock, we order ALL we need for the next year, and THEN I make a footprint and put it in the design. For EVERY SINGLE PART. Starting with the IC's. It actually works quite nicely once you get used to it. Obviously, there are some losses there too - just the cost of doing business in these crazy times.
I can't even get cables anymore. Or connectors. It's an insane situation, and the company I work for isn't built to manage this level of churn in our products. How do you support customers when equipment BOMs change every week? We just can't keep up.
7nm or whatever state-of-the art processes may be important for certain latest electronics, but I'm guessing there are many components that could use 10 year-old or more semiconductor fabrication processes.
If you're willing to share the original part number, I'm curious as to what FET could cause so much grief.
Same with USB-UART bridges; zip, nada, nothing. I found some Cypress parts a few weeks ago, and I should consider myself lucky.
I won't order PCB until I have all reels of parts on my desk.
If you wire up a solar cell like an LED, it glows dimly in infrared. QA uses this to diagnose dysfunctional wafers.
A gigawatt of solar cells represents about 5 square kilometers of silicon wafers at 20% light conversion efficiency. The world installed 183 gigawatts of solar PV in 2021, almost all of it based on silicon wafers:
https://www.pv-magazine.com/2022/02/01/bloombergnef-says-glo...
That's in the neighborhood of 915 square kilometers of wafers.
Silicon for solar has risen meteorically over the past 20 years.
https://www.pv-magazine.com/2021/10/26/whats-next-for-polysi...
Until the early 2000s, demand for polysilicon (often simply referred to as “poly”) was dominated by the semiconductor industry, which required a fairly steady 20,000 to 25,000 metric tons (MT) per year. But semiconductor demand for poly was quickly outpaced by PV as the solar industry began to grow rapidly, from a rounding error at the turn of the millennium to almost half of global polysilicon demand by the middle of the decade.
...
By the end of 2013, the manufacturing cost of polysilicon had tumbled to below $20/kg among industry leaders. Meanwhile, capacity had grown from less than 50,000 MT per year in 2007 to over 350,000 MT per year by 2013.
Polysilicon capacity at the end of 2021 was in the neighborhood of 700,000 metric tons, with more big expansions on the way. The extra 350,000 metric tons added since 2013 is almost entirely for solar.
MEMS. Micro-electromagnetic systems. The most common MEMS I can think of is the comb sensor, used for accelerometers in all of your cell phones.
https://www.memsjournal.com/2010/12/motion-sensing-in-the-ip...
The MEMS sensor for an accelerometer is quite simple. Take the nearest comb and smack it against a desk: you'll notice that the comb vibrates in one direction. Now hook up two combs and interleave their teeth together so that they're barely touching. When they touch, an electrical signal is sent through them to sense when they touch.
Add differently sized teeth, the larger the spacing the more acceleration is needed before they activate. (EDIT: Looks like the iPhone MEMS uses capacitance... similar concept though, the capacitance changes based off of how far away these teeth are from each other and you can measure that using college-level electronics)
Finally, have these teeth rotated in all directions, so that you can sense all the directions in one little device.
--------
MEMS are about using the physical properties of object, but just making these small physical objects really, really, really tiny thanks to the magic of photolithography.
You can see this literal comb structure by looking at any accelerometer under a microscope: https://memsjournal.typepad.com/.a/6a00d8345225f869e20148c70...
------
If the accelerometer is too difficult for you to understand, the "beginner MEMS" is gears.
https://www.sandia.gov/app/uploads/sites/145/2021/11/1-1.jpg
You can make any shape you want with modern chip-making tools. The "shape" most people want is a transistor (gate, drain, source). But in many ways, a teeny-tiny gear is simpler to think about.
The practical applications of micro-scale MEMS (gears, combs, springs, etc. etc. ) is somehow harder to think about than computers, so there aren't very many practical MEMS around. But still, practical MEMS help remind us that all of these chip-making tools exist in the real, physical world. Albeit at a very small scale.
E.g. EPROM (memory type chip) is typically deleted by shining a uv light on the actual silicon die, through a uv transparent quartz window in the final packaged chip.
edit: fixed EEPROM -> EPROM
Also, EPROMs are extremely vintage. They were replaced by EEPROMs - the first of which came out in 1977. That's an... extremely vintage example.
most recently: https://dyson-sphere-program.fandom.com/wiki/Microcrystallin...
https://youtube.com/watch?v=BM7VDOoFIWI
The LED is the component on the left; there's a very dim flash (pretty much just the black die turning red) at around 11s, then every few seconds.
I can't remember exactly how it works... I think there's two capacitors charged up to the voltage of the LED in parallel through high-value resistors, and a circuit that shorts the +ve of one to the +ve of the other to put them in parallel.
It only just works at a very specific light level. IIRC some of the transistors are used as very low leakage diodes rather than transistors, as the regular diodes I had we're too leaky.
In the solar world they call it a blocking diode.
- hard drive read/write heads (the platters are debatable)
- inkjet printer nozzles (this is why making a DIY inkjet printer is nontrivial)
- air pressure sensors (e.g., for car tires)
- precise frequency filters for smartphone wireless communication
- oscillators (https://news.ycombinator.com/item?id=18340693)
- very tiny microphones for smartphones (speakers are harder)
- Digital Micromirror Devices (DMDs): arrays of tiny mirrors used in most projectors
- microfluidics ("lab-on-a-chip" stuff for fast disease testing, DNA sequencing, cell manipulation, etc)
And a couple other semiconductor applications:
- LCD/LED screens (monitors, phones, laptops, etc) (these are made on a glass surface instead of a silicon wafer but use the same basic manufacturing techniques)
- laser diodes (laser pointers, CD / Blu-ray players)
- many quantum computers
Yeah I remember a friend of mine back around '90 wanted to try it out. I can't remember if it was using DRAM or EPROM memory though. I want to say it was called 'ramra' or 'ramera'.
This should be called a Goku circuit. Kaaaamaaaaikaaaaaa.
For some of the other "jelly bean" parts we need (o-rings, snap rings, etc) we are looking at making them in-house, but both the raw materials and machinery to make them are not possible to get. We could spend the next few months making our own machines to fabricate them, but without being able to source steel and various rubbers any more reliably than the finished goods, there isn't much point.
At this point, its tempting to try and raise capital to start mining and smelting aluminum, steel and buy an oil well and small refinery so we can ensure we have the materials needed to keep production smooth.
It's also gotten to the point where there doesn't exist enough stock in distribution to buy a year's worth. And I'm not talking high volume, maybe 1k/year to 50k/year. Distributors are constantly decommitting from orders, broker stock is drying up, etc
We've spent several hundred thousand at "Win Source" broker in China, and haven't had a problem yet (knock on wood). We X-ray and test to verify though.
Mostly, Chinese brokers are a den of thieves/a pool of sharks. If they can counterfeit it, they do. Use a credit card to help with clawing your money back in case of fraud. And never ever buy IC's from Amazon or Ebay. Those are ALL fake.
Don't you find that they can be resold? Surely there's someone desperately looking for the parts you find unusable, but would fit their needs...
New Craigslist category for semiconductors?
Except that now there isn't enough capacity to go around.
Or you could buy a lithium-ion battery off the shelf today at 95% round trip efficiency and low self discharge.
[1] https://en.wikipedia.org/wiki/Luminous_efficacy#Lighting_eff...
Isn't that one of the possible uses for superconductors tho?
IIRC, EPROMs were still cheaper than EEPROMs for many years. EPROMs probably were sold in commercial quantities well into the 80s, and maybe used in the 90s.
EPROMs were erased by just throwing them into a UV-bin and blasting them with UV light. In contrast, EEPROMs needed transistors inside to handle the erasing cycle.
Finally, EPROM's last stand was as a low-cost one-time-programmable ROM (aka: PROM). All you had to do was make the same chip except without the expensive "window" (that'd normally receive UV-light for erasure).
IIRC, students recently graduated when I was doing my freshman year had used them for projects in their freshman year, but not since - so they were probably commonplace until 1990 or so. At least in Trondheim, Norway.
Industrial equipment moves at a different pace, and in these days of 10x price jumps and "52 week" lead times, sometimes dusty relics from the 80s wind up being relevant in 2022!
I had professors who were active during the time and gave me the rundown. I have touched EPROMs and all that good stuff, still part of the labs at my college and my professor liked talking about "the good ol days". But I've never in fact used them in any practical manner.
That's vintage, isn't it? In the 90's, stuff from the 60's was 'vintage'. By 2022... the 90's are vintage?
And thanks for the info, I don't have direct experience dealing with EPROM and I had conflated EEPROM and EPROM together in my mind, but a quick google search quickly reveals my inadequate knowledge, which has now been updated, even if only good enough for trivia.
Another piece to think about is density. Even with these losses.. how much energy can a material hold in terms of pure light? Is there a limit to how much light can pass through a material?
And I wonder if you could slow the light down even further, maybe 2 or 3 more orders of magnitude..
That being said, I have to agree that the word 'vintage' evokes the 50-70s more than the 90s. Maybe the word has just come to mean that era in the english language, just like 'olden days' tends to mean a pre/semi-industrialized era that's somewhat locked in time.
Perhaps we're running out of words for the past and we simply call these eras by their decades now.
