IBM's Transistor Data Revealed

Atryn writes "After last week's story — Intel and IBM both announcing breakthroughs in chip design enabling continued adherence to Moore's Law — many folks wondered how and why the two companies' announcements came out simultaneously. The Register explains it, and as a bonus they are releasing a leaked copy of IBM's future research documentation (PDF)."

Silicon Valley will become K-Valley then?

[namityadav]

From Intel: High-K Material is a material that can replace silicon dioxide as a gate dielectric. It has good insulating properties and also creates high capacitance (hence the term "high-k") between the gate and the channel. Both of these are desirable properties for high performance transistors. "k" (actually the Greek letter kappa) is an engineering term for the ability of a material to hold electric charge. Think of a sponge. It can hold a lot of water. Wood can hold some but not as much. Glass can't hold any at all. Similarly, some materials can store charge better than others, hence have a higher "k" value. Also, because high-k materials can be thicker than silicon dioxide, while retaining the same desirable properties, they greatly reduce leakage.

It had to happen eventually...

[Anonymous Coward]

...Otter got something *right*!

\Nobody likes a braggart.

Re:Silicon Valley will become K-Valley then?

[exley]

No. The substrate that these chips are fabbed on is still silicon. The article is somewhat misleading because it's the gate oxide, which is typically made of silicon dioxide, that's being replaced with the hafnium-based high-k material. I'm loathe to site this as a source but since it has pictures, here [wikipedia.org] is a Wikipedia article that will show you the basic structure for anyone unfamiliar.

I also find it interesting that they are using metal gates instead of polysilicon, considering that metal gates were used in the olden days before the switch to poly.

Re:Silicon Valley will become K-Valley then?

[exley]

Despite the fact that you decided to go troll on us with how you responded to the parent post, I'll bite since this was something I got to thinking about as well.

Higher gate capacitance means that you can get more charge in the channel for a given gate voltage (Q=CV). This can give both higer currents and a reduced threshold voltage, which are good things.

But higher capacitance, of course, slows things down when you get to thinking about those RC time constants. So, do the benefits of higher capacitance outweigh the negatives? IBM and Intel seem to think so.

Re:Silicon Valley will become K-Valley then?

[Umbrel]

The improvement is not about increased capacitance in each transistor channel, that would be bad. The capacitance is scpecifically increased in the gate, that means that the gates can be made thicker (less leakage currents = less power consumption) while keeping (or improving) the values for current and voltaje needed to be applied at the gate and the time for the transistor to switch.

Re:Silicon Valley will become K-Valley then?

[Anonymous Coward]

"I also find it interesting that they are using metal gates instead of polysilicon, considering that metal gates were used in the olden days before the switch to poly."

Yup, we used to use metal gates, but the difference in work function between the gate and the silicon caused problems. It gets somewhat confusing, but I'll try to explain briefly. The work function is a basic property of a metal or semiconductor that indicates the distance from the vacuum level to the fermi level of a plain piece of that material. When two materials with different work functions are placed in contact, the fermi levels align, and this can "bend" the conduction bands on the semiconductor. In the operation of a MOSFET these bands are also bent in various ways by voltage applied to the gate, but if the original bend is too large, it would take a huge voltage to move the conduction band to where its needed. This can result in a threshold (or turn on) voltage that is too large to be usable.

When transistors were first developed, there were metals that had somewhat decent work functions, but they weren't great, and since its a basic property of the metal, it was a fixed value, that gave a pretty firm threshold voltage. As the process was scaled down the threshold voltage didn't scale with it. Then someone came up with the idea of using polysilicon for the gates, which solved all the work function problems. Polysilicon is just silicon grown on top of the wafer without the stringent growth conditions, seed crystals, controls, etc. that produce a wafer made out of a single crystal, so it is formed of many small crystals of silicon instead of one. But, since its still silicon, the work function is close to that of the rest of the transistor, and can be adjusted and fine tuned to any work function needed by doping it along with everything else.

However, while polysilicon gates solved the work function problem, they created new problems of their own. They aren't as conductive as a metal, so you get some resistive loss in the gate and an increased delay when coupled with the gate capacitance. You also have a small depletion layer that forms against the oxide when the transistor is turned on, simmilar to the channel that forms on the other side of the oxide. This depletion region effectively adds extra thickness to the oxide, and thus lowers its capacitance. Lowered cap at that point in the transistor means less control over the channel, and reduced performance. You can thin the oxide to make up for it, but that gives you more leakage (the depletion region in the poly will conduct if the oxide leaks anything, so it doesn't help with leakage). High-K materials can also help with this issue, but they can only do so much.

When the switch to polysilicon gates was first started the added resistance was pretty much negligable since the gates were so large, and the depletion region was much smaller than the oxide thickness so it wasn't the dominant factor in gate capacitance. After much scaling, however, the gate resistance is now signifigant and the loss of gate cap is pretty bad. Companies have been working for quite some time to find a metal that would work to replace poly, but its been difficult to find an alloy with a work fuction close to silicon, or that can be adjusted to be close to silicon. Its finally been done, so the next generation or two of processes will see the return of metal gates.

Metal has always been better, its just been too difficult to use metal once the polysilicon trick was developed, but the polysilicon "hack" is starting to break down, so we're forced to go back to metal gates and find some way to make it "work".

Re:Silicon Valley will become K-Valley then?

[exley]

The gate itself isn't inherently capacitive, it's the capacitance of the gate/oxide/channel structure that we're talking about, so you can't decouple gate capacitance and channel capacitance. Charge on the gate (which you can think of as the top plate of a parallel plate capacitor) results in equal but opposite charge on the channel (the bottom plate, with the insulating oxide acting as the dielectric).

You are absolutely correct that the point in increasing the thickness of that oxide is to reduce leakage and power consumption. But increasing the thickness of the gate oxide lowers the capacitance of the structure, so they want high-k materials to compensate for this.