Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?
This is your opportunity to ask questions. No question too simple or too silly.
Culture war topics are accepted, and proposals for a better intro post are appreciated.
Jump in the discussion.
No email address required.
Notes -
Does anybody have any questions about working in a semiconductor fab is like? Is there a market for writing up an effortpost on how semiconductor manufacturing equipment works? I met a handful of techy people this weekend who were fascinated by it and asked non-stop questions, so I figured there may be some interest here.
I'm very interested and would enthusiastically read as much as you're willing to write about the subject.
More options
Context Copy link
I always assumed the career prospects weren’t that good since you often end up getting tied to one highly specialised machine/procedure of one company and then they have no incentive to pay you. Is this correct?
Career prospects are better than you describe and the specialization actually helps. Most companies use the same baseline tools with exceptions existing for high-end tools (see EUV in my other reply) and what we call "legacy" tools, or extremely old tools that aren't sold anymore and OEM and second-hand support availability is minimal. For example, almost all companies own a fleet of AMAT Endura tools. So if I were to work at company A for 10 years specializing on Enduras I could easily transfer over to company B and work on theirs. Even if your exact toolset isn't there, the principles stay the same across the board, enough so that you can make up for lack of experience and knowledge quickly.
Experience is also a major consideration. Where I think intelligence reigns supreme in more "theoretical" roles (research scientist, low-level chip designer, etc), experience is king for equipment and process engineers. Intelligence helps and there's definitely a minimum requirement, but you don't have to be a genius to create extensive personal or company-wide documentation on how your tools work, understand major events that had a long troubleshooting process, come up with improvement projects, or run basic process experiments. By this, the longer I stay working on these tools the more I see, the more I learn, and the more pieces I can connect together to make improvements. I can then jump ship and immediately start contributing to another company, especially if I've uncovered or implemented things they haven't yet.
Regarding actual pay, one of my colleagues got a 40% raise by moving from company C to company D. COL was the same. He already had 20 years of experience at company C!
More options
Context Copy link
I would presume that this depends greatly on what role you have. Research scientist? Engineer? You have plenty of flexibility and good pay, albeit places like TSMC have an abysmal work culture. I'd expect that the kind of technicians who are skilled enough to work in a fab have options too, and can re-skill.
More options
Context Copy link
More options
Context Copy link
I'd really like to hear about the day to day procedures, clean room process and whatnot. And an overview of stuff like how new your machines are, what's needed to switch to making different chips, etc
Daily work varies wildly. My title is equipment engineer, also known as tool owner (equipment is interchangeable with tools), so I'm responsible for making sure certain pieces of equipment (about 25 in total, which is a decent amount of anybody) in the fab are running properly. We also have 24/7 equipment technicians that will fix the tools. As my old boss described it, my job is to make sure the tools don't stop working, while the technicians want to get them working again if they're broken. When they're all working well or the techs are handling it themselves, I work on self-conceived projects to make them run better, faster, or longer. The manufacturing environment can be a bit brutal since I'm technically on call 24/7 for issues the technicians can't handle or aren't involved in, which requires the occasional weekend laptop log-in.
So what are some things I've done over the past few weeks? (Apologies for the vagueness)
Cleanroom protocol is surprisingly lax compared to cutting-edge fabs. There's no air shower to remove particles from the bunny suit, nobody freaking out that your nose is out, and plenty of dirty-ish parts and hand tools lying around all over the place. This is allowed because a) wafers are almost always contained in their own mini environments, whether it's inside the tool or inside their carrier (called a FOUP and pronounced foo-p), and b) our technology node size is a bit larger and a few particles here and there isn't catastrophic.
You'd be surprised how old our equipment is. Semi equipment is notoriously expensive, so when you combine that with a company that is notoriously cheap and processes that don't require the best equipment on the market you get some old equipment that we're just forced to take care of. Plus if ain't broke, don't buy a new one. A few of my tools are almost 20 years old now, and Theseus doesn't own them—many still have original parts on them! Thankfully the OEM still does a decent job of a) offering spare parts to support part failures, and b) offers replacement parts for obsolete parts. My newest tool was manufactured in 2019. The fab regularly installs new tools as we remove old ones and ramp our production levels.
Preventive maintenance is critical to ensuring parts on the tool last a long time (like how your engine lasts as long as your oil) and preventing product from scrapping because the process' tolerances are all out of whack.
Different chips generally means smaller chips, which require more advanced tools, especially in the photolithography department (also called photo or litho for short). I think this video, this video, and this series offer an excellent overview of cutting edge litho methods that are required to manufacture low nm nodes you hear about coming from Intel, TSMC, Samsung, etc. It's important to note the insane capital required not to just invest in a fab itself, not just the tool that go inside, but the litho tools themselves. New SOTA EUV tools cost around $200MM, or over 1% of a (higher end) fab's cost, and that's just a single tool. Ouch!
How many processes depend on gravity at all? How many require specifically 9.8 m/s^2?
More options
Context Copy link
Thanks for taking the time, this is fantastic. A couple of questions, I'm sure I'll have more if you're willing to write more:
What kind of inputs (like industrial gases, chemicals, etc.) go into the tools? Are they connected with fixed pipe or are they manually loaded in?
How is the entire end-to-end process controlled at the fab? As in, is there a centralized control room that oversees the entire operation, or is it understood that the individual tool stations run as close to 100% utilization as possible?
What kind of redundancy is built-in to the process? For example, if a tool went down, does it cause everything behind it to grind to a halt, or are there tools in parallel? Can wafers be stored until they need to be used, or do they have a short "shelf life"?
Do you know where your finished products are used? (i.e. do they go into cars? consumer electronics? defense tech? communications? etc)
I think we have a very similar professional role. I'm manufacturing-adjacent (oil & gas, electrical engineer working in maintenance of power systems) and had an idea of doing a similar type of Q&A. Because of the nature of the industry there aren't a lot of people who have both firsthand knowledge and are able to publicly write about it. There was an interesting book I read a while back from a former controls engineer called "A Funny Thing Happened On The Way To The Control Room" which might be of interest to you or anyone else following this thread. It's unfortunate that there's so many interesting anecdotes that can only be disclosed after one retires.
Edit: It looks like there's a similar version of that text here: https://www.emersonautomationexperts.com/wp-content/uploads/2024/04/Process-Control-Case-Histories-Greg-McMillan.pdf
Happy to answer any other questions. I really enjoy talking about this and find it equally as cool.
Fabs use a combination of bulk gases (N2, H2, O2, Ar, etc) and speciality gases (AsH3, PH3, SiH4, SiH2Cl2, etc) depending on the process. Bulk gases are fed from massive canisters and get distributed throughout the fab to points of use. The piping is normally located directly underneath the main fab floor in an area called the subfab to save space, increase convenience of maintenance, and prevent particles from contaminating tools. Speciality gases follow the same path from their source canister, but instead there are valve manifold boxes (VMBs) between the point of use and source to allow for safer operation and improved monitoring capabilities. MKS has a decent fab facilities overview here. (As a side note, welding gas lines is preferable to minimize the chance of leaks or contamination. This comes at the risk of the line being completely custom and having long leadtimes in case it needs to be replaced. I prefer parts to be as modular as possible so we can replace the part itself and not the entire subsystem with it.)
The process is controlled by the brains of the fab, the manufacturing execution system (MES). Some fabs build their own custom MESs to match their needs and others go with out-of-the-box solutions that have dedicated company support. Full-stack MESs generally handle most of the calculations when decided what to do, whereas not-full-stack MESs require other programs to assist.
Redundancy is crucial to a fab's success. We try to minimize OAK (one-of-a-kind) paths else everything grinds to a halt directly in front of that tool and I get yelled at for why my tool isn't up. Industrial engineers are able to model a fab's capacity abilities and determine how much of what technology is able to run given the number of available tools and their qualification status. For example, I have four tools (E1-4) and four technologies (T1-4). E1 can run T1-4, E2 can run T2 and T4, E3 can run T3, and E4 can run T1 and T4. Thus, T1 has two paths, T2 has two paths, T3 has one path (OAK alert!), and T4 has three paths. T4 material would likely be fine since it has three different options to run through. T3's OAK is a bit dangerous and unOAKing it should be a priority if its loadings (how much T3 we run) is high enough. To put it more simply, think of it as tolls: if there are 10 lanes and 10 consecutive tolls (so 100 stations total) and all of set1's tolls can handle Toyotas, but only one of set2's tolls can handle Toyotas, then Toyotas will get through set1 quickly but get really backed up at set2 because they're all forced to the same path that has a fixed throughput and may be dealing with other car brands! Some wafers require processing within a certain amount of time after finishing their previous process for various reasons (e.g., native oxides).
I will kindly abstain from answering this for opsec reasons :)
Thanks for the excellent response!
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link