Talk:Semiconductor device fabrication

Untitled
Please define exactly what 'feature size' refers to; there too many vague definitions widely differring from companies to schools. tks. — Preceding unsigned comment added by 88.218.215.231 (talk) 06:01, 17 October 2012 (UTC)

Comment 1
The oxidation link leads to Redox which is not appropriate for semiconductor device fabrication because it ignores the growth of an SiO2 film on top of a silicon substrate. Propose breaking out "Oxidation" from Redox in order to address this issue and talk more specifically about silicon reactions with oxygen and why such reactions are performed. Alternatively, if someone has objections to this use of "Oxidation", a topic such as "Oxide Growth" or "Oxide Formation" (which could also include oxide deposition) could be used. Amadeust 21:27, 12 Nov 2004 (UTC)

IMPORTANT add 14nm process
With Intel and IBM/ARM having already announced fab plants under construction, also AMD has clearly indicated 14nm chips as part of their roadmap, the relevant links:

Intel = http://www.geek.com/articles/chips/intel-building-14nm-chip-facility-in-arizona-for-5-billion-20110221/

IBM/ARM = http://www.pcadvisor.co.uk/news/index.cfm?newsid=3257351

AMD = http://www.brightsideofnews.com/news/2010/11/10/amd-moves-cpu-architecture-into-tick-tock-mode3b-schedules-20nm2c-14nm-apus.aspx

the article should be updated to include this information, it's not really my field, so i'll leave it to someone else.

JavaByte (talk) 06:37, 25 February 2011 (UTC)

Wafer cost questioned
I have issues with this part: "In 2005, a typical cost for a processed 300mm wafer was about $US 300,000. The desire for profit inspires manufacturers to pack more devices on a wafer."

The total surface area of a round 300mm-diameter wafer is about 71,000 mm2 (square millimeters, don't know how to superscript.) Using my Sempron CPU as an example, it uses about 100mm2. So ignoring waste about 700 Semprons-to-be fit on a wafer. That's over $400 each, and we're a long, long way from a finished product. I'd expect my $70 Sempron to cost at least $2000 if the $300,000 wafer cost is correct.

Also, they cut 1000-2000 wafers off an ingot these days, implying PER INGOT production cost of $300,000,000 to $600,000,000 if $300,000 is correct for a single wafer.

Does anyone have a better number for wafer cost? Shyland 23:58, 15 April 2006 (UTC)

Update: I'm editing the section about wafer cost based on data from icknowledge.com. Estimated cost for a 300mm wafer for 130nm IC's is currently around $2500-$3500. Shyland 02:34, 16 April 2006 (UTC)

I seem to remember from a couple of year's ago Discovery program that a wafer costs around $8000 (at that time obviously). There's no way it can cost $300,000 these days. A few thousand dollars is a good estimate, although I'm sure the exact price varies a lot with process size, number of masks, etc, etc. Romanski 21:53, 15 May 2006 (UTC)

The cost of the processed cassette holding 25 wafers could be close to the figure of $300,000. Moreover, its an opportunity cost of the entire business chain from designers to the final retailer of the chip. The cost at the time of final processing (Just before slicing & dicing) may not be near that —The preceding unsigned comment was added by 203.91.207.30 (talk • contribs).

Comment: The major initialy outlay for producing a wafer is the lithographic mask production costs. These are typically over $1million for 130nm and below, how you relate this to the wafer cost depends upon how many wafers you produce. Hence people typically only produce IC's if their volumes are very large. I can't comment to the exact cost of a wafer or the per wafer processing costs once you have the masks, but this is certainly much much cheaper than the initial outlay.

"A number is" vs. "a number are"
I disagree with the last edit to change "a number are" to "a number is". Grammar authorities generally call for collective nouns indicative of a plurality of something to take the plural verb. Accordingly, "a number are" is correct and it certainly pleases my ear more. Still, I don't want to be the one to just edit it back to the previous version.


 * Perhaps you have the wrong article? I can't find that phrase in this article. Aaronmz 01:52, May 12, 2005 (UTC)

The recent content deletions
I'm always troubled when an anonymous editor comes along and deletes a bunch of apparently-valid content from an article (as just happened here). Ordinarily, I'd revert the change but in this case, some value was added as well. Does anyone have any opinions on the accuracy or importance of the deleted 'graphs:


 * It is important for these workers to minimize their exposure to the toxic materials used in manufacture, such as arsenic; for this reason, the semiconductor fabrication facilities are highly automated.


 * Memory chips are typically the first devices to use a technology, as they are highly regular, and can be used to test a technology.


 * In an effort to increase profits, semiconductor device manufacture spread from Texas and California in the 1960s to the rest of the world, such as Malaysia and Japan, and is a global business today. Some Malaysian facilities are now in their fourth decade of operation, for example.

(Some of that last 'graph now exists in the recent edit.)


 * Microprocessor manufacture is still strong in the United States, but the world's market leader is in Taiwan, represented by major companies such as Taiwan Semiconductor Manufacturing Company and United Microelectronics Corporation (see UMC web site). Singapore is the second biggest semiconductor manufacturer in the world, where Chartered Semiconductor Manufacturing dominates.

Opinions?

Atlant 16:41, 24 May 2005 (UTC)
 * I reinstated the Hazard notice. The information about RAM is in the IC article. The history of semiconductor manufacture in Malaysia etc might be reinstated. But it may well be the case that these fabs are long gone. Ancheta Wis 11:54, 30 May 2005 (UTC)

---

My comments on the deleted sections - I would say they needed revising at least.

The reason for increased automation in semiconductor fabrication facilities (fabs) is not a safety measure, though safety may improve as a consequence. Automation reduces the chance of "mis-processing" material which can result in "scrap". Human operators make mistakes while a programmed automated process, when correctly set up, does not.

With regard to RAM leading the way I would say that memory and microprocessor technology are increasingly taking different paths. Memory devices perhaps rely more on novel memory cell designs, on ways to reduce process steps and dealing with extreme physical features such as trench depth. Microprocessors, in the hunt for process speed, lead the way in minimum device sizes, new engineered substrates (such Silicon on Insulator and Strained Germanium Silicon) and new interconnect techniques an material (copper, dual damascene, etc).

As for the spread of manufacturing, the desire to build IC's in the US is fading fast and Taiwan, now the king, will probably hand its crown to China in the coming decades, at least with respect to numbers. Malaysia has been more involved in relatively low-tech post-fab dicing, bonding and packaging operations and is not an IC fab stronghold. Interestingly the most advanced, mass-produced microprocessors are still produced in the "West", especialy in the USA and Europe.

Supplied by Rutherford - 13 June 2005


 * I've rewitten the hazards section - before reading this discussion (blush). But I think what I've written takes account of what was being said above.  I can confirm that we used to use all that hazardous stuff before there was significant automation - the automation was put in to improve reproduceabilty mainly.  --Phil Holmes 14:09, 6 September 2005 (UTC)


 * Phil,


 * We use to seperate acids and solvents too.
 * Acids went into a tank and solvents went down the drain.
 * May want to label in the article the same?

Cheers

Scott 15:25:22, 2005-09-06 (UTC)

As measured by sales, the largest semiconductor manufacturers in the world are Intel, Samsung, and Texas Instruments. (Data from IC Insights.) It's hard to compare TSMC and other foundries, since foundries don't sell directly to end users, but the claim that Singapore is the second biggest manufacturer in the world is, um, questionable at best. Edited the section to reflect this.--Katherine (Was registered user, but lost ID info.) 66.30.210.131 21:47, 9 February 2006 (UTC)

Correct the source links
There's an extensive list of articles that link to Fabrication (semiconductor). Now that that is a redirect page, we should edit all the source links to correctly link to the real article here at Semiconductor fabrication.

Atlant 17:17, 5 August 2005 (UTC)

Front end/Back end?
The article refers to front end processes (photolith and doping, etc.) and back end processes (metal, etc.). It's been a while since I worked in the industry (mid 80's) but we never referred to these processes as front or back end. Has anyone else ever heard of them? --Phil Holmes 14:12, 6 September 2005 (UTC)

Front end and Back end refer to the use of Copper (Cu) in the processs. FE is usually to about the 1st metal layer and all of that is done without Cu because of its migratory properties and hazards to the chemistries in the early stages of the process. Once the interconnecting layers are started (they are Cu now) then the process becomes BE. These terms are mainly for segregation of tools so that there is no Cu/Non-Cu cross contamination.

Device Test & Tungsten
In my opinion, in this section should be a link to the DFT page and some description about that: http://en.wikipedia.org/wiki/Design_For_Test Opinions?

And a Question about the vias: I recently learnt that the lower metal layers (1&2) are connected with tungsten vias (called tungsten-plugs) to the silicon and amongst each other. In the higher metal layers, the vias are the same material as the metal. Why is tungsten used for metal 1&2?

Andreas


 * At least one reason tungsten is used is because when the metal layers are copper, the copper would contaminate the semiconductor. The tungsten acts as a barrier to copper contamination. For aluminum metalization, I don't know; someone else will have to answer; my next guess would be for better compatibility between the coeffients of thermal expansion.


 * Atlant 20:23, 9 February 2006 (UTC)

Tungsten also offers better electromigration resistance (though worse electrical resistance) than aluminum, so tungsten vias can block electromigration. -- Katherine 66.30.210.131 21:59, 9 February 2006 (UTC)

I have seen some images (transistor crossection) and additionally to your thoughts I think to have two additional ones: 1. It seems you can (due to some reasons I don't know) achieve way better aspect ratios with tungsten. All the aluminum interconnects on higher levels had considerably worse aspect ratios and showed intense underetching (.18 CMOS). 2. Maybe there are some high temperature processes to do while the first via layer is already applied. Eutectic Si-Al melts at 577°C which is not very much for some diffusion processes and stuff.

The thermal expansion seems to be a good point as well. Thermal expansion coefficients: Si: 2.6 µm/(m·K) W:  4.5 µm/(m·K) Al: 23.1 µm/(m·K)

Andreas


 * I'd like to note that choice of metal for interconnects and vias varies from process to process. This may seem obvious, but Andreas's original comment seems to imply otherwise.
 * I'd guess that it's a high-temperature issue. Even short of melting and alloying, aluminium and silicon have a huge problem with spiking.  I can explain, if you'd like.
 * I doubt that thermal expansion is a problem. Thermal expansion creates huge stresses at interfaces with thermal oxides, but people usually don't mind.
 * --Smack (talk) 02:04, 14 June 2007 (UTC)

May I ask to what does the sentence "This takes six to eight weeks" in the "Device test" section refer to? Laser etching, chip mounting, or the entire testing process? If it confused me, you can be sure it will confuse a few more people. A little bit of disambiguation by someone more knowledgeable would be welcome, i.e. make it clearer. -- xompanthy 18:45, 11 March 2006 (UTC)

Perhaps the entire beginning-to-end IC production process? I'm mystified too. Shyland 02:58, 16 April 2006 (UTC)

Requested move
I think this article should be entitled "Semiconductor device fabrication". It mostly seems to address IC product fabrication and doesn't really discuss any of the material system development issues that the current article title might imply. In addition, semiconductors themselves aren't "fabricated" strictly speaking, but refined grown into crystals, and processed for electronic/electromechanical device construction. If there are no objections, I'll move the article soon. -- mattb


 * Also consider "microelectronic fabrication". --Smack (talk) 18:34, 18 January 2007 (UTC)


 * No, as this would exclude MEMS, which are still mostly a semiconductor field (with exceptions). It is also not a very suitable title for devices like LED, lasers, APDs, etc, which are generally called "optoelectronics."  I think "semiconductor device fabrication" is an appropriate title. -- mattb
 * Move done as requested. DES (talk) 18:59, 6 April 2007 (UTC)

Merge
This article overlaps greatly with Microfabrication. I admit that the two terms are not exactly synonymous, and each includes some areas that the other does not. However, I think that we should restructure our coverage of micro/electronic fabrication to eliminate this overlap and clearly identify non-overlapping areas. --Smack (talk) 18:34, 18 January 2007 (UTC)


 * Strong oppose. Semiconductor fabrication is a big enough topic without merging in all the non-semiconductor microfabrication stuff.  If there's too much overlap, take some of the semiconductor part out of microfabrication. Dicklyon 18:48, 18 January 2007 (UTC)
 * Oppose per Dick. While microfabrication has largely been a semiconductor-dominated field in the past, the development of biomedical MEMS and the like have provided the motivation for developing microfabrication techniques for totally unrelated materials like polymers.  These articles should remain separate and have little overlap, even if the articles' current status don't reflect this ideal. -- mattb
 * Close for lack of support; leave separate articles. Dicklyon 18:22, 21 January 2007 (UTC)


 * But we still have overlap. Would someone like to remove it? --Smack (talk) 19:11, 21 January 2007 (UTC)


 * Add a comment to the microfabrication talk page to suggest what work it needs; some overlap is OK, but that article shouldn't need to get into the details that this one covers well. Dicklyon 19:16, 21 January 2007 (UTC)

RCA clean
I wasn't sure what to do with this article submission, so I declined it. If it is notable, please feel free to merge the content in here or create a new article. Articles for creation/2006-04-15. davidwr/ (talk)/(contribs)/(e-mail)  23:42, 13 August 2007 (UTC)

Good article
http://www.pcplus.co.uk/node/3059/ —Preceding unsigned comment added by 93.173.239.146 (talk) 07:37, 23 May 2009 (UTC)

Front-end vs. Back-end
The article states there are two "steps" -- front-end processing and back-end processing. It then has a sub-heading for front-end processing, but does not have a back-end processing heading. It should clarify. Drcwright (talk) 01:22, 8 September 2009 (UTC)


 * I did a bit of restructuring to clarify. It could use more work. Dicklyon (talk) 23:21, 26 September 2009 (UTC)

Aluminum spelling
John changed aluminum to aluminium per the guideline at Manual_of_Style_(chemistry). It strikes me as odd, as in the semiconductor industry I've always seen it as aluminum. Opinions? Dicklyon (talk) 23:17, 26 September 2009 (UTC)
 * I am the editor who amended the spelling. As an international project, Wikipedia sometimes needs to make compromises over matters of national varieties of English. For elements we use the IUPAC spellings aluminium, caesium and sulfur on science articles, even though national spelling variants may use aluminum, cesium or sulphur. See also Aluminium. Hope that makes sense. --John (talk) 03:10, 27 September 2009 (UTC)


 * Makes perfect sense, but the guideline says there can be exceptions, and this is a technology article more than a science article, so I thought it would be best to seek a consensus of editors with respect to whether it should be treated according to the science article guideline or otherwise. Dicklyon (talk) 03:18, 27 September 2009 (UTC)


 * Fair point. I'll obviously go along with any consensus generated here. --John (talk) 03:24, 27 September 2009 (UTC)

Missing manufacturing processes
In the table with manufacuring processes, 500nm, 150nm, 80nm, 55nm and 40nm are missing.

There could be others missing, but these are the ones I know of. Examples for these processes are;

500nm: 3dfx Voodoo Graphics & Rush chipsets.

150nm: ATi/AMD's Radeon 8500 & 9xxx series

80nm: ATi/AMD's Radeon X1550 & X1650

55nm: ATi/AMD's Radeon HD 3xxx & HD 4xxx series

40nm: ATi/AMD's Radeon HD 5xxx series.

NitroX infinity (talk) 12:44, 5 June 2010 (UTC)

There are also some memory and flash processes, which has other steps of minimal feature size `a5b (talk) 20:25, 6 June 2010 (UTC)

merge older processes into the article
Most articles about manuf. processes (350 nm, 500 nm and so on), especially old, are very short. It will be better to move older processes into the article. `a5b (talk) 20:24, 6 June 2010 (UTC)

BEOL and FEOL articles
There are stub-class BEOL and FEOL articles (Backend-of-the-line processing and Frontend-of-the-line processing). I added some books to them and some information too. But they need a lot of work. `a5b (talk) 00:08, 17 November 2010 (UTC)

Restructured See Also section
OK, I have had a go at restructuring the See Also section. What I have done is put in subheadings, one for each of the headings in the main article; but I have deleted and modified a few, where appropriate (for example, the list of process steps in the main article is really already a See Also section in its own right for wafer processing and die processing).

This does lead to one possible idea... of absorbing each of my proposed subsections of the See Also section into their respective places in the main article. I will let someone else do that, if it is considered preferable.

For my part, I think that it might be good to encourage this See Also list to grow (and hence, my proposed structure, so that it grows in a structured way). At present it is far from an exhaustive list. On the other hand, the reader can probably find most of the missing topics by clicking on each of the main topic links, and then looking at their respective See also sections.

I note that there was a discussion on this talk page, some time ago, on the justification for this article, the choice of name for the article, and the decision not to merge it with a related article. This is all important stuff, of course, when considering how people wish to develop this article from here.

Just my thoughts. Feel free to build on, and rip down, what I have proposed (of course! it goes without saying, I suppose) or to contact me, if you would like me to do something further to add to this article. TheAMmollusc (talk) 07:26, 16 June 2011 (UTC)

Timeline is off
I know this sounds naive, but isn't the timeline really off? It is currently middle of 2012 as I write this comment, but 22nm has only recently release by intel. No other major firms have 22nm products on mass production. The timeline labels 22nm as 2011... — Preceding unsigned comment added by 128.12.232.145 (talk) 06:21, 23 May 2012 (UTC)

Process Technology
I think there should be a separate article, or a section of this article that is called "Process Technology". It should be something like the article at:

http://lookup.computerlanguage.com/host_app/search?cid=C999999&term=process+technology&lookup.x=-410&lookup.y=-693

To the layman, without this separate and distinct explanation of "process technology", it is hard to read Wikipedia and get a clear understanding of the meaning of the list of line sizes (32nm, 22nm, etc) that is shown in the upper-right corner of this and other articles. The use of the term "node" in its sense of a step along the path of process technology, should also be explained. --Westwind273 (talk) 19:21, 18 April 2013 (UTC)

"Semiconductor node"
"Semiconductor node" redirects here. Please explain the concept. Staszek Lem (talk) 19:49, 19 January 2016 (UTC)
 * It now redirects to ; does that explain the concept well enough? Guy Harris (talk) 07:07, 23 May 2022 (UTC)

Moving from 2-years manufacturing time frame to 3-years manufacturing time frame
Intel has recently anounced that it leaves the ‘Tick-Tock’ 2-years manufacturing phase to a 3-years manufacturing phase - ‘Process-Architecture-Optimization’. So it should be clearly predicted, acording to what we know these days, that from the year 2017 - the year that 10 nm process manufacture will go out the markets - every ~3 years the manufacturing of the chips should shrink by 2, insted of every 2 years by now. So therfore, the 7 nm manufacturing process should arrive in ~2020, and 5 nm manufacturing process shoud arrive in ~2023. 85.64.251.242 (talk) 11:22, 27 April 2016 (UTC)


 * Very interesting, do you have a source/link for that? Is it just Intel or also ITRS is changing its projected timeline? --Nanite (talk) 15:57, 27 April 2016 (UTC)
 * http://www.extremetech.com/extreme/225353-intel-formally-kills-its-tick-tock-approach-to-processor-development
 * http://www.theinquirer.net/inquirer/news/2452294/intels-tick-runs-out-of-tock
 * http://arstechnica.com/information-technology/2016/03/intel-retires-tick-tock-development-model-extending-the-life-of-each-process/


 * I don't know about ITRS. But I guess it's very much related to Intel's plans and abilities. 85.64.251.242 (talk) 22:07, 27 April 2016 (UTC)


 * OK, I'm not sure where if that belongs on this article, since presently the two year node cycle isn't mentioned. But it's worth a mention on some other ones like Die shrink and 10 nanometer -- go ahead and be WP:BOLD. :) --Nanite (talk) 16:07, 28 April 2016 (UTC)

"Ink in the water" story? Real or legend?
A 2020 Wall Street Journal op-ed by Andy Kessler [|China is losing its bet on chips] mentioned an incident where an engineer accidentally spilled ink into a fab's water supply, and yields actually went up:

"a Silicon Valley engineer once accidentally spilled ink into a fab’s water supply—and output yields went up!"

Is this real, or it just one of those apocryphal stories that floats around that nobody is sure about? If it is, does someone know where I could find more information about it - and *why* it improved things? Jimw338 (talk) 02:26, 26 November 2020 (UTC)

Source 2 does not validate claims
Source 2 does not mention anywhere in its article how long it takes to fabricate a silicon wafer. It is about water consumption but the largest fabs in China. 198.161.136.29 (talk) 05:33, 17 December 2021 (UTC)

Some technology nodes are missing
TSMC offers nodes 28nm, 16nm, 12nm but they aren't listed on the page. 2610:1C1:1:6074:0:0:16:84 (talk) 22:07, 16 March 2022 (UTC)