F.Y.I...Advanced 0.13-[micro]m CMOS Process Is Ready To Produce 2-GHz Processors.

[Quiet109]

Two major semiconductor players have unwrapped plans to ramp up production in the 0.13-[micro]m (130-nm) CMOS process. Combining innovations in copper wiring, silicon-on-insulator (SOI) transistors, and improved low-k dielectric insulation, IBM will begin manufacturing for external customers this year. Meanwhile, Intel is migrating to the 0.13-[micro]m CMOS process for achieving speeds beyond 2 GHz for future Pentium lines.

Intel says it will begin producing its 130-nm CMOS process on 200-mm wafers this year. Slowly, the company will migrate to 300-mm wafers next year. In its 130-nm process, Intel has reported 70-nm gate lengths and 1.5-nm gate oxide thicknesses (see the figure). The 130-nm process features six layers of dual damascene copper interconnects while using a high aspect ratio (thickness/width) of 1.6:1 to obtain high circuit density and lower line resistance simultaneously. These ultra-small transistor features were demonstrated using normal deep-ultraviolet lithography tools.

The manufacturer's researchers also have made strides in further scaling the transistor size down to 30 nm with a gate-oxide thickness of only three atomic layers. As a result of this breakthrough, Intel will be able to build microprocessors containing 400 million transistors, running at 10-GHz speeds and operating at less than 1 V, within the next five to 10 years.

"As our researchers venture into uncharted areas beyond the previously expected limits of scaling silicon, they find Moore's law still intact," says Sunlin Chou, vice president and general manager of Intel's Technology and Manufacturing Group. In scaling the device size to a new low, Intel's researchers have addressed the electrical leakage problem that has been a barrier.

"Our research proves that these small transistors behave in the same way as today's devices, thereby ensuring that there are no fundamental barriers to producing these devices in high volume in the future," notes Gerald Marcyk, director of Intel's Components Research Lab. Technology and Manufacturing Group.

Unlike Intel, IBM's 130-nm SOI process implements up to nine layers of copper wiring. Labeled CMOS 9S, it employs the SiLk dielectric with a k value of 2.7. This technology will be used to produce high-performance microprocessors with docking speeds over 2 GHz and high-density, high-speed SRAMs as well.

IBM's Semiconductor Research and Development Center, East Fishkill, N.Y., will serve as a pilot production line for CMOS 9S. The company then will move the 0.13-[micro]m process to its high-volume Burlington, Vt., facility this year.

This process was developed jointly with Infineon Technologies AG of Germany and foundry service provider United Microelectronics Corp. (UMC) of Taiwan. Infineon and UMC also are designing and building chips with the 0.13-[micro]m CMOS process. Next, the three firms plan to move toward 0.1-[micro]m feature sizes.

[Tac]

WOW.

Guess ill be jumping from 300mhz to 2.5 ghz next year      

[bloom25]

Truth be told silicon semiconductor technology is getting pretty close to it's absolute limits.  Cmos transistors have a relatively large capacitance at high frequencies.  This reduces their gain and creates a large amount of wasted energy.  It also delays the switching times.

For those of you who are curious, here are the equations for the maximum possible frequency attainable by a cmos transistor before leakage becomes too large:

fT=gm/(2pi*(Cgs+Cgd))  gm is what is known as the transconductance and is equal to sqrt(2*Kp*I*W/L) where I is the current through the device.  Kp is a constant determined by the manufacturing process (see the article above) and W is the width and L is the length of the device.  Cgs is gate to source capacitance and Cgd is gate to drain capacitance.  One obvious way to increase maximum switching frequencies is to increase the width of the device and/or reduce the length.  (This is what they have done above.)

SOI (silicon on insulator) is another way to reduce capacitances.  CMOS transistors on a chip in the past have all shared a common "body" which is grounded for Nmos transistors and connected to V+ for Pmos tranys.  When you then turn on the device by applying the opposite voltage to the gate you are basically creating a capacitor as well.  By insulating the body of the transistor you in essence allow the body to "float" reducing the voltage potential and hence the capacitance.  (It's pretty easy to see this if I could draw it for you.)

I'll try:  Nmos transistor:
gate ->---------
______________________
|_n__|____p____|__n__|
drain           source
terminal        terminal
(The insulator would go about here.)

(I hope the formatting doesn't get messed up.)  Basically a cmos (be it Nmos or Pmos) transistor works like a capacitor.  In the case of an n-mos (n-channel metal oxide semiconductor) the p+ section would be an entire layer of the chip.  The n areas are then created by "shooting" impurities to create the p layer into the correct locations.  (This is done with group 7 elements.)  You then put a small oxide layer over the p and then (originally) a metal plate over that.  (Now they use poly silicon, which kind of makes the name inaccurate.)  To activate the device you apply a small positive voltage to the gate which in effect "pulls" electrons toward the top of the p layer.  This creates a channel where a current can flow from one n part to the other.  Unfortuately this also is a parallel plate capacitor.  At high frequencies the resistance of a capacitor gets very small, thus high leakage currents.

You might have noticed just how thin Intel is now claiming the oxide layer is.  (3 atoms!)  You can't make it much thinner.  

Within the next decade or so something new will need to come along or else we will be stuck as far as clock speeds go.  Not only are we running into a size barrier, but the purity of the silcon required in creating a device that is literally only a few atoms thick itself will become cost prohibative.

I don't know of any other practical alternatives to Cmos transistors at this time as far as silicon goes.  BJT's can switch at even higher speeds, but they also draw much more current.  They are also somewhat harder to design with IMO.

Well hopefully I haven't bored you too much.  This is what I get to play with everyday.  

Edit: It messed up my picture alright.  The n terminals are the source and drain respectively.  (Actually they are identical and interchangable.)

------------------
bloom25
THUNDERBIRDS

[This message has been edited by bloom25 (edited 03-05-2001).]

[AKDejaVu]

Bloom25,

One thing the microchip manufacturing companies have done successfully over the last 6 years is prove scientists wrong.  I was there when the "its impossible to define structures smaller than the wavelength of light being used to expose them" rule was blown away.  I'll probably be not too far away when someone says "wow.. they said a transistor could never switch this fast".

I do believe 10GHz will be doable in the next 7 years.  I really wish I could tell you why.

AKDejaVu
Components Research: Intel Corperation

[Mk10 225th]

Deja, please keep up your wonderful and important work.

I have watched my retirement (INTC), go down the tubes since November 7, and have taken out a personal loan to do bills until some checks start coming in, as opposed to selling my dilapidated Intel stock.

It is nice to hear that there is a possibility that behind closed doors in Silicon Valley, men and women in dark robes, veiled in secrecy, are doing everything they can to make sure I can retire before age 80.

If you ever feel disillusioned about your work, bored, or that no one cares, just think about the idiots like me out there, who are your silent cheering section, urging you on to victory and success.

I imagine you to be an intelligent, handsome individual, with a winning sense of humor, and in general, a credit to the human race as a whole.

If you ever need any cheering up, words of encouragement, or a cold beer if you're in mid-Missouri, just give me a shout.

I love you man.

Mk

[Maverick]

Please..... Someone translate the previous very learned posts into layman's terms!!!!!!!!    

Mav

[bloom25]

AKDeja VU, do you think it would be possible to make the gate oxide layer any thinner than 1 atomic layer?    (Unless you guys find a way to split atoms on a chip soon.   )

I suppose you could make the transistor wider and the length shorter, but that would make it take up more space.

I don't think that the wavelength of the light is an issue at this time, but I do think that CMOS transistors are reaching their limits.

(There has been some interesting research on creating very small BJT transistors, but that is probably 10 years away from being practical.  Basically they used a "thread" of p type and doped the ends as n type and created an NPN trany.  BJTs can switch faster than CMOS transistors, but power will increase.)



------------------
bloom25
THUNDERBIRDS

[Tac]

What's this about? Quantum Computer?
 http://www.zdnet.com/zdnn/stories/news/0,4586,2615525,00.html?chkpt=zdhpnews01

"NEW YORK -- IBM Corp. said on Tuesday it had developed the world's most advanced quantum computer, a device based on the mysterious quantum physics properties of atoms that allow them to work together as a computer's processor and memory.
IBM (NYSE: IBM) said the computer, which uses five atoms to work as its processor and memory, demonstrates for the first time the potential of such devices to solve certain problems at a rate remarkably faster than conventional computers. The experimental machine is considered the next step toward a new class of devices capable of superfast calculations.

"A quantum computer could eventually be used for practical purposes such as database searches -- for example, searching the Web could be sped up a great deal -- but probably not for more mundane tasks such as word processing," said Isaac Chuang, the IBM researcher who led the team of scientists from IBM, Stanford University and the University of Calgary.

A quantum computer could also be used for cryptography, or the making and breaking of codes. This has drawn the interest of the U.S. National Security Administration and the Department of Defense, which are funding Stanford's efforts to build the quantum computer.

Where Moore's Law ends The current method of creating processors, which are becoming increasingly smaller and more powerful as described by an axiom known as Moore's Law, is expected to reach a barrier sometime in the next decade or so. This process, lithography, will not allow for the creation of microchips the size of molecules, prompting researchers to try to build computers by using genetic strands or developing other tiny technologies.

"Quantum computing begins where Moore's Law ends -- about the year 2020, when circuit features are predicted to be the size of atoms and molecules," said Chuang. "Indeed, the basic elements of quantum computers are atoms and molecules."

Chuang said in an interview that his team used the test quantum computer to solve a typical mathematical problem used in cryptography -- finding the period of a function. The computer was able to solve any example of the problem in one step, while a conventional computer would require repeated cycles to solve the problem.

The road to quantum computing Chuang said the experiment showed the viability of the quantum computer.

"I think this experiment shows we are on a pathway which is predictable and understandable, that quantum computers will be useful someday," he said.

The quantum computer is based on the spin of an electron or atomic nucleus, and the strange properties of quantum particles in which they can spin simultaneously in different directions if they are not observed.

When the spin of a particle is up, the atom can be read as a one, and the spin down can be read as a zero, corresponding to the digital ones and zeros that form the binary language of traditional computers. Such devices use transistors, which are turned on and off to represent the ones and zeros.

On-off, up-down, yes-no? What makes quantum computers unique, however, is that quantum particles can also be in a state of "superposition" -- spinning simultaneously up and down.

"Due to their small size and if they are very well isolated, they can be spinning up and down at the same time," said Chuang.

This state would represent both zero and one and everything in between. Instead of solving the problem by adding all the numbers in order, a quantum computer would add all the numbers at the same time.

This phenomenon permits a quantum computer to have enormous power, Chuang said. For certain types of calculations, like complex algorithms for cryptography or searches, a quantum computer using several hundred more atoms in tandem would be able to perform billions of calculations at the same time.

However, it is unclear when such a computer would be commercially available. Chuang said it is expected that between seven and 10 atoms will be used in tandem in more advanced quantum computers within the next two years."


[This message has been edited by Tac (edited 03-06-2001).]

[AKDejaVu]

Just in the news today:

 http://public.wsj.com/sn/y/SB983982441220068847.html

If the buyout of SVGL (Silicon Valley Group Lithography) by ASM (A Dutch company) gets blocked, < .13um lithography will be made more difficult.

I somewhat understand the concerns... mostly if SVGL is bought out, the US does not have a really high-end stepper manufacturer any more.  They will be owned by Nikon and ASM.  Our big semi-conductor companies are AMAT and Novellus.. neither of wich make steppers.

The sad thing is, its not really technology based.  ASM has recently done things that make most other stepper companies seem like they are opperating in the dark ages.  They already have the technology, they are just building resources.

Oh well, we are eagerly awaiting the outcome of this one.. I can assure you.

AKDejaVu