Intel's Tri-Gate gamble: It's now or never

Deep dive into Chipzilla's last chance at the low end


Deep – really deep – inside Tri-Gate

As we mentioned earlier, Tri-Gate is a variation of the FinFET concept, in which instead of lying below the gate as in a traditional planar transistor design, the charge-carrying silicon pokes up into the gate itself.

Intel electron-microscope photo: planer 32nm versus tri-gate 22nm

In a planar design, gates were lain across a flat silicon substrate – not 3-D enough for marketeers, however.

This non-planar construction, by the way, is why you'll hear Intel marketeers exalting the Tri-Gate design as "3-D". Well, yeah, that's not entirely untrue, but the gate-wrapped fin is hardly 3-D enough to warrant such buzz-lines as this one from the Wednesday presentation: "Transistors have now entered the third dimension! – and, yes, that's Intel's italics and Intel's exclamation point.

In any case, you might reasonably ask: "So the fin sticks up in the gate? So %$#@!ing what?" Bohr's answer to that perfectly reasonable question is equally reasonable: "The key advantage of this structure – and it comes from the fact that the gate wraps around this fin – [is that] it provides fully depleted operation of the transistor."

Bohr's answer, of course, is only reasonable if you know why a fully depleted transistor is a good thing – which takes a bit of background to explain.

When a traditional microprocessor transistor is in the "on" state, current flows from source to drain through what's called the transistor's "inversion layer", which interacts in Intel's designs with the transistor's high-k metal oxide interface with the gate. That's all well and good.

However, when the transistor is in the "off" state, some charge can still trickle through the silicon substrate. When it does, the transistor is regarded as not being fully depleted. That's bad.

When voltage hangs around like this, it degrades what chip boffins call the "sub-threshold slope" of the transistor – which Bohr defined as "essentially the transistor turn-off characteristics," or which we can think of as how "off" the off state actually is.

When the transistor's off state is fully depleted, it helps minimize power usage. Also, a fully depleted transistor can have a lower threshold voltage – meaning that the voltage needed to switch it from off to on can be lower, again saving power.

"When you operate at a lower threshold voltage," Bohr explains, "you have improved performance. Or you can operate at lower voltage, and if you operate at lower voltage, there's a significant active-power saving – and that's really probably the most important advantage of Tri-Gate."

So, a fully depleted transistor both leaks less power and requires less power. All good.

Planar processors are hell to make fully depleted. In their traditional state, there's a lot of room in the silicon substrate to house errant voltages. You can, however, add an oxide insulator below the source and the drain to create what's called a partially depleted silicon-on-insulator (PDSOI) design.

You can go all the way to fully depleted (FDSOI) without going FinFET by depositing an extremely thin SOI layer on top of the oxide – but reaching full depletion this way is quite expensive. According to Bohr, baking a chip this way adds at least 10 per cent to the total wafer cost.

Tri-Gate, by comparison, is cheap. "Another nice thing about the Tri-Gate devices," Bohr says, "is that they're not that expensive to add. Compared to a planar version of 22 nanometers, Tri-Gate transistors add only about 2 to 3 per cent cost to the finished wafer."

The Tri-Gate way of reaching full depletion is to stick that silicon fin up into the gate, have the inversion layer on both sides and the top of the fin, with the high-k metal oxide of the gate snug against the inversion layer. Presto – no room for nasty voltages to accumulate, plus the larger wrap-around inversion layer and metal-oxide interface allow for more current, and thus better performance. Bonus!

Another nifty Tri-Gate trick is that you can have multiple fins in the same transistor: more fins, more current, better performance. Or, as Bohr puts it succinctly: "You can gang together these fins to create larger transistors with higher drive currents."

This "How many fins do you want?" capability will help Tri-Gate to populate throughout Intel's product line, from low-power single-fin transistors to multi-fin structures that will find their way into Xeons – and even into Itaniums, when that benighted chip reaches its Kittson incarnation, two generations from today's 9300 series.

Broader topics


Other stories you might like

  • Former AMD chip architect says it was wrong to can Arm project
    Cancelled the K12 CPU? Big mistake. Huge, says Jim Keller

    Amid the renewed interest in Arm-based servers, it is easy to forget that one company with experience in building server platforms actually brought to market its own Arm-based processor before apparently losing interest: AMD.

    Now it has emerged that Jim Keller, a key architect who worked on Arm development at AMD, reckons the chipmaker was wrong to halt the project after he left the company in 2016.

    Keller was speaking at an event in April, and gave a talk on the "Future of Compute", but the remarks were unreported until picked up by WCCF TECH.

    Continue reading
  • Intel says Sapphire Rapids CPU delay will help AMD catch up
    Our window to have leading server chips again is narrowing, exec admits

    While Intel has bagged Nvidia as a marquee customer for its next-generation Xeon Scalable processor, the x86 giant has admitted that a broader rollout of the server chip has been delayed to later this year.

    Sandra Rivera, Intel's datacenter boss, confirmed the delay of the Xeon processor, code-named Sapphire Rapids, in a Tuesday panel discussion at the BofA Securities 2022 Global Technology Conference. Earlier that day at the same event, Nvidia's CEO disclosed that the GPU giant would use Sapphire Rapids, and not AMD's upcoming Genoa chip, for its flagship DGX H100 system, a reversal from its last-generation machine.

    Intel has been hyping up Sapphire Rapids as a next-generation Xeon CPU that will help the chipmaker become more competitive after falling behind AMD in technology over the past few years. In fact, Intel hopes it will beat AMD's next-generation Epyc chip, Genoa, to the market with industry-first support for new technologies such as DDR5, PCIe Gen 5 and Compute Express Link.

    Continue reading
  • AMD bests Intel in cloud CPU performance study
    Overall price-performance in Big 3 hyperscalers a dead heat, says CockroachDB

    AMD's processors have come out on top in terms of cloud CPU performance across AWS, Microsoft Azure, and Google Cloud Platform, according to a recently published study.

    The multi-core x86-64 microprocessors Milan and Rome and beat Intel Cascade Lake and Ice Lake instances in tests of performance in the three most popular cloud providers, research from database company CockroachDB found.

    Using the CoreMark version 1.0 benchmark – which can be limited to run on a single vCPU or execute workloads on multiple vCPUs – the researchers showed AMD's Milan processors outperformed those of Intel in many cases, and at worst statistically tied with Intel's latest-gen Ice Lake processors across both the OLTP and CPU benchmarks.

    Continue reading

Biting the hand that feeds IT © 1998–2022