Home

News

Forums

Hardware

CPUs

Mainboards

Video

Guides

CPU Prices

Memory Prices

Shop



Sharky Extreme :

 
Search
 



Latest News


- Outdoor Life: Panasonic Puts 3G Wireless Into Rugged Notebooks
- Averatec Launches Lightweight Turion 64 X2 Laptop
- Acer Fires Up Two New Ferrari Notebooks
- Belkin Debuts Docking Station for ExpressCard-Equipped Notebooks
- Logitech 5.1 Speaker System Puts Your Ears At Eye Level
News Archives

Features

- SharkyExtreme.com: Interview with ATI's Terry Makedon
- SharkyExtreme.com: Interview with Seagate's Joni Clark
- Half-Life 2 Review
- DOOM 3 Review
- Unreal Tournament 2004 Review

Buyer's Guides

- September High-end Gaming PC Buyer's Guide
- September Value Gaming PC Buyer's Guide
- October Extreme Gaming PC Buyer's Guide

HARDWARE

  • CPUs


  • Motherboards

    - Gigabyte GA-965P-DS3 Motherboard Review
    - DFI LANPARTY UT nF4 Ultra-D Motherboard Review

  • Video Cards

    - Gigabyte GeForce 7600 GT 256MB Review
    - ASUS EN7900GT TOP 256MB Review
    - ASUS EN7600GT Silent 256MB Review
    - Biostar GeForce 7900 GT 256MB Review





  • SharkyForums.Com - Print: Presentation of Intel and 20nm transistor

    Presentation of Intel and 20nm transistor
    By Arcadian June 18, 2001, 05:29 PM

    For those of you that don't know, Intel recently announced that they have been able to manufacture a transistor whose gate length is only 20nm in size. This is only months after announcing the world's first 30nm gate based transistor, which eclipsed IBM's earlier announcement last year of a 50nm gate length transistor. To put that in perspective, current microprocessors are on a 180nm process with gate lengths as small as 130nm, and Intel's future process on which the Tualatin and Northwood will be introducing a 130nm process with gate lengths as small as 70nm.

    Now, Intel has released a presentation that shows just what it means to be able to produce smaller and smaller transistors.
    http://www.intel.com/research/silicon/20nmpressfoils.pdf

    Some things I found the most interesting (and there are many) is that in 2003, Intel will be launching a 90nm process. Originally, we had all assumed it would be 100nm, but now it looks like Intel has set higher goals than previously thought. Likewise, other future processes have been subsequently scaled smaller, showing a clear path until the end of the decade.

    Other pieces of this presentation show Intel's results when compared to other published results, showing them with a sizable lead. Also, photos of the silicon gate have shown it to be very accurately constructed. This is very good news, as it appears that the silicon medium will last at least until the end of the decade before something else is required to move on.

    I welcome any comments to this. What do other people think?

    By sww June 18, 2001, 09:59 PM

    Reading this, along with some other things I've been reading has me wondering:

    1. Is the voltage of the chip the voltage necessary for a gate switch and this is why voltage goes down with a new process?

    2. How did Intel go from having transistors and gate lengths of the same size to the gate lengths being smaller?

    3. If #1 is correct, why does AMD need higher voltages for the same size gates? Are they using the same process size, but having to use longer gates like Intel used to?

    If all of this is utter nonsense, feel free to tell me and point me somewhere that I can update my knowledge.

    By pm June 18, 2001, 11:25 PM

    quote:Originally posted by sww:
    1. Is the voltage of the chip the voltage necessary for a gate switch and this is why voltage goes down with a new process?
    Back in the good old days - like the Pentium 60 & 66 and previous, the industry used to do something called "constant voltage scaling". Scaling refers to the fact that the process used would scale by a factor (the lowercase Greek symbol alpha is usually used), while voltage would remain constant at 5V. But this ran into problems, and so from 5V we dropped to 3.3V back around 0.35um, and then from 3.3V down to ~2.2V at 0.25um and then onto ~1.7V at 0.18um.

    In a transistor, there are several things which are scaled by 'alpha' (usually ~30%) when you shrink a process, the channel length (L) which is the distance across the transistor (and should, in my mind be called the width, but let's not go there), the channel width (W) which is the length of the transistor itself, the voltage (V), the oxide thickness (tox), and, more confusingly, a bunch of things relating to the way the transistor is created (like the junction depth).

    If you scale all these, then the properties of the transistor change such that everything works pretty well. More relevant to your question is that by reducing the gate oxide thickness, and reducing the voltage, the electric field across the gate remains constant - this allows the device to remain fairly reliable (ie. it will last a long time) and keeps most of the transistor properties pretty linear (which means that you can take a device from an older/larger process, shrink it, and still have it work fairly well on the new one with very few changes.

    But, if, as was the case for a long time, you keep voltage (V) constant and shrink everything else, then the electric field across the transistor gate increases by 'alpha'. This is great if you want really fast transistors (because, as every good OC'er knows, higher voltage means faster CPUs), but it has two really bad effects which eventually caught up with the industry: heat and gate-oxide breakdown (due to various wear-out mechanisms such as hot-e, etc).

    Heat on a CPU is generally defined by a wide variety of factors but the basic formula is P=Cap*freq*(voltage)^2. So, if you shrink a transistor, cap goes down, the transistors are faster so freq will go up, and if voltage is constant then power goes up by a square factor. Anyone who remembers the original 5V 60MHz and 66MHz Pentiums will remember that, for their time, they were hot CPUs. It was clear to most that voltage was due for a drop.

    Reliability is a more subtle concern, but it is of key importance to the semiconductor industry. Reliability drops dramatically with increased voltage due to the effects of the voltage itself and also to the increase in temperature. So if you want long lastly transistors, the voltage needs to drop. For a long time, the industry was able to ignore this, but eventually electronics caught up with everyone and since then the voltage has dropped at approx. the same rate as everything else.

    quote:2. How did Intel go from having transistors and gate lengths of the same size to the gate lengths being smaller? I don't understand what you are saying here.

    quote:3. If #1 is correct, why does AMD need higher voltages for the same size gates? Are they using the same process size, but having to use longer gates like Intel used to? There are two choices if you want to raise the voltage of a CPU on a given process (which thus increases the switching performance of the transistors): You can tweak a process to allow higher voltages or you can allow a higher voltage at the cost of decreased reliability. I'm sure that AMD has done the former, and not the latter. All 0.18um processes are not the same... in fact, TSMC (a Taiwanese foundry) has 3 different 0.18um processes that they advertise and two of these have varying voltages and two of these are set to a higher voltage than AMD is currently using (IIRC, they are at 1.75V). So, what one company's process voltage is set to doesn't necessarily mean that another one is setting theirs too high or too low... there are a lot of parts to the fab recipe and everyone's is a little different.
    http://www.tsmc.com.tw/technology/index_cl018.html

    By pm June 18, 2001, 11:36 PM

    quote:How did Intel go from having transistors and gate lengths of the same size to the gate lengths being smaller? I was thinking about this... are you referring to the fact that "effective" transistor length is less than the drawn transistor length? By this I mean that companies (not just Intel, pretty much everyone does this) will claim that their 0.18um transistors have channel lengths of 0.13um or less?

    By Arcadian June 19, 2001, 01:53 AM

    quote:Originally posted by pm:
    I was thinking about this... are you referring to the fact that "effective" transistor length is less than the drawn transistor length? By this I mean that companies (not just Intel, pretty much everyone does this) will claim that their 0.18um transistors have channel lengths of 0.13um or less?

    Patrick, if you look at my link above, and then go to page 7, you will see the past and future processes that Intel intends to produce. On every process past P854, Intel used smaller gate lengths than the actual lithography design rules. Sww was asking why this is. I am kind of curious myslef. Besides the range of this particular statistic, I am not fully sure of the purpose.

    By chickenlump2001 June 19, 2001, 10:00 AM

    I think he was talking about how a .18u press will create 180nm transistors, and .13u press creating 130nm transistors, etc etc.

    How did they create a 20nm transistor with existing technology?

    quote:Originally posted by pm:
    [QUOTE][b]How did Intel go from having transistors and gate lengths of the same size to the gate lengths being smaller? I was thinking about this... are you referring to the fact that "effective" transistor length is less than the drawn transistor length? By this I mean that companies (not just Intel, pretty much everyone does this) will claim that their 0.18um transistors have channel lengths of 0.13um or less?

    [/B][/QUOTE]

    By Arcadian June 19, 2001, 10:22 AM

    quote:Originally posted by chickenlump2001:
    How did they create a 20nm transistor with existing technology?

    I'm guessing that they were using prototype EUV lithography tools, which provide an order of magnitude better resolution for etching out transistor features within the silicon wafer.

    By pm June 19, 2001, 11:22 AM

    So, there are three possible interpretations of question #2 that I have seen:

    1. How can you have a transistor channel length that is substantially less than the actual drawn gate length? In other words, why do 0.18um transistors have channel lengths that are much smaller than 0.18um? (pm's interpretation)

    2. How do companies use light to draw detailed features on a chip when the wavelength used is much higher than the thing that you are trying to define using it. For example, a lot of 0.18um (180nm) processes use 248nm light to define the features, how does this work? (Arcadian's interpretation)

    3. How do companies manage to continue to shrink process technology to smaller and smaller levels? (Chickenlump's interpretation)


    Any and all of these are excellent questions.

    #2 is the subject of a huge post that I have been gradually putting together over the last week talking about deep-UV, optical proximity correction (OPC) and phase-shifted masks. I'm running into problems writing it due to a lack of time (we are moving houses right now) and the fact that this 'article' will need pictures to make it truly understandable. I intend to write an Arcadian-length post on the subject in the near future, but I may need to put it on a website due to the pictures problem.

    SWW, which one were you asking?

    By sww June 19, 2001, 07:14 PM

    My question #2 is:

    With 350 nm lithography, the gate length is 350 nm. With 250 nm, the gate length is 200 nm. When you go to 180 nm transistors, the gate length narrows (shortens) to 130 nm. And at every point thereafter, the gate length is shorter than the lithography size. Why?

    By pm June 19, 2001, 11:39 PM

    quote:Originally posted by sww:
    My question #2 is:

    With 350 nm lithography, the gate length is 350 nm. With 250 nm, the gate length is 200 nm. When you go to 180 nm transistors, the gate length narrows (shortens) to 130 nm. And at every point thereafter, the gate length is shorter than the lithography size. Why?

    "Why" is the shorter answer - the longer question is "how". So, my fingers thank you.

    Really fine lithography relies on four major things: a very good light source, very high quality optics, the materials used to make the mask, and the light-reactive material (the 'photoresist'). It turns out that there are either very few 'points' along the wavelengths where all four of these come together well.

    157nm
    193nm - ArF
    248nm - KrF (also known as deep-UV)
    365nm (also known as 'i-line' litho)
    436nm (also known as 'g-line' litho)

    Why these particular ones? Usually the light source determined the wavelength. But the rest of it is required too - you need have a material for the masks and optics that won't absorb too much light, and another material that can be laid accurately down which will be completely opaque (to define the 'dark' zones) and you need a good photoresist for that wavelength. So, when all these aligned then you have a process technology. It has gotten trickier (it seems) the lower down the industry has gone. At 193nm there aren't a lot of materials tranparent to the light so you have to deal with degrees of opaquacy (word?). If the optics and the mask absorb too much light, they'd get hot which would distort the optics, and there wouldn't be as much energy getting through to the photoresist. So they play with materials.

    Here's a great article on the subject that I found while trying to figure out which wavelength correlated to i-line and g-line. This is very readable (IMO) and covers the fascinating (again, IMO) subject of OPC. This is pretty much the article that I've been fiddling with writing, although I discuss phase-shifting as well and this one ignores that interesting technique.
    http://www.fabtech.org/features/lithography/articles/9.205.php3

    Edit:

    Hmm. I was poking around fabtech.org (the link above) and found this article which answers the question more thoroughly than I did. http://www.semiconductorfabtech.com/features/lithography/articles/edition9/ed9_a3_1.shtml

    By Arcadian June 20, 2001, 01:04 AM

    quote:Originally posted by pm:
    157nm
    193nm - ArF
    248nm - KrF (also known as deep-UV)
    365nm (also known as 'i-line' litho)
    436nm (also known as 'g-line' litho)

    i-line and g-line. Would these perhaps stand for indigo and green wavelengths of light? Just curious.

    By pm June 20, 2001, 01:47 AM

    quote:Originally posted by Arcadian:
    i-line and g-line. Would these perhaps stand for indigo and green wavelengths of light? Just curious.
    Perhaps... I'm not sure of the origin of the terms - in fact I couldn't even remember the wavelengths. IIRC, there's also the much older h-line and k-line which I couldn't guess how to translate.

    I don't think it has to do with colors directly - I think it has something to do with spectrum emission lines. IIRC, astronomers use similar verbage when they are doing spectroscopy of stars. But it's late and I couldn't find any proof to back up my guess in my quick websearch. I'm curious too.

    By sww June 20, 2001, 08:42 PM

    First G-line with = 436 nm, then I-line with = 365 nm.

    Yes.


    quote:Originally posted by Arcadian:
    i-line and g-line. Would these perhaps stand for indigo and green wavelengths of light? Just curious.


    Contact Us | www.SharkyExtreme.com

    Copyright © 1999, 2000, 2001 INT Media Group, Incorporated. All Rights Reserved.


    Ultimate Bulletin Board 5.46




    HardwareCentral
    Compare products, prices, and stores at Hardware Central!


    Copyright © 2002 INT Media Group, Incorporated. All Rights Reserved. About INT Media Group | Press Releases | Privacy Policy | Career Opportunities