The semiconductor industry is about to experience a tectonic shift. Wave goodbye to traditional transistors based on polysilicon-gate electrodes and silicon-dioxide (SiO2) dielectric insulators, which had been used to make the transistor gate dielectric for more than 40 years because of its manufacturability and ability to deliver continued transistor performance improvements. Say hello to hafnium-gate dielectric insulators and new non-disclosed metal materials for the NMOS and PMOS transistors that make up CMOS semiconductors (Fig. 1).
Intel is leading this change with the first working 45-nm multicore processors using these novel technologies. The company plans on selling CPUs that use the high-k + metal-gate (HK+MG) stackup in the second half of this year (Fig. 2). Based on the new Penryn die, these CPUs have been tested with the Windows Vista, Mac OS X, and Linux operating systems (Fig. 3).
"Intel has developed a complete high-k plus metal-gate solution, with a high-k gate dielectric and two types of metal gate electrodes optimized for NMOS and PMOS transistors," says Kaizad Mistry, 45-nm Program Manager, Logic Technology Development of the Technology and Manufacturing Group for Intel. Not since 1969 has there been such a dramatic change to transistors.
"These materials have been integrated together to form a reliable and manufacturable 45-nm process technology, which allowed us to demonstrate the world's first working 45-nm CPUs with high-k plus metal-gate transistors just a couple of \[months\] ago," adds Mistry. "We believe no other company has achieved this level of success with these new transistors." Intel also indicated it has five early version Penryn-based products up and running, and the company plans to build 15 45-nm processors total.
"The implementation of high-k and metal gate materials marks the biggest change in transistor technology since the introduction of polysilicon-gate MOS transistors in the late 1960s," says Intel co-founder Gordon Moore. Compared to its 65-nm counterparts, the 45-nm HK+MG stackup offers higher transistor density, faster switching speed, reduced gate and source-to-drain leakage current, and reduced switching power (see the table).
Along with its research partners AMD, Sony, and Toshiba, IBM announced a similar breakthrough, with plans to use an HK+MG transistor in a server-based IC in 2008. Not to be outdone, Sematech also announced an HK+MG transistor stack. The bandwagon jumping will continue as other fabs follow with their own flavors of HK+MG solutions.
Intel believes these other companies will need a year or two to catch up. In the meantime, creating circuits using 45-nm process technology is still feasible without the combined use of the new materials. However, leakage would be too problematic at 32 nm and below using the older poly plus SiO2 stackup.
Silencing critics who said it couldn't be done while delivering density increases and reductions in leakage currents, these new technologies are helping Intel, IBM, and other companies revive Moore's Law for a few more generations. Intel expects to be at 32 nm in 2009 and 22 nm in 2011.
WHAT'S ALL THIS HIGH-K STUFF, ANYHOW?
Gate leakage current, source-to-drain leakage current, and increased heat dissipation have plagued the industry as transistors have scaled down in size. In 2004, Intel found that leakage power accounted for 25% of total chip power, and this number continues to increase with each new process shrinkage.
"As more and more transistors are packed onto a single piece of silicon, the industry continues to research current leakage reduction solutions," said Mark Bohr, Intel senior fellow. "Meanwhile our engineers and designers have achieved a remarkable accomplishment that ensures the leadership of Intel products and innovation. Our implementation of novel high-k and metal gate transistors for our 45-nm process technology will help Intel deliver even faster, more energy-efficient multicore products that build upon our successful Intel Core 2 and Xeon family of processors and extend Moore's Law well into the next decade."
Gate leakage occurs when electrons leak across the very dielectric barrier intended to keep them in place. Source-to-drain leakage takes place when the transistor is supposed to be in the "off" state, yet current still flows from source to drain as it would in the "on" state.
The term "high-k" implies that the material maintains a high dielectric constant (with respect to SiO2) that creates a high field effect between the gate and the channel. Also, it features great electronic insulation properties. The "k" refers to the material's ability to hold electric charge.
When evaluating the performance of two dielectric materials (hafnium and SiO2 in this case), the relative physical thickness required to produce the same gate capacitance can be compared, providing an idea of relative "electrical thickness." This is commonly known as the equivalent oxide thickness (EOT). An EOT of 1 nm would result from using a 10-nm thick dielectric featuring a k value of 39. Comparably, the k value of SiO2 is 3.9.
Metal gates also help increase the gate field effect. Together, the HK+MG stackup helps to significantly reduce leakage while increasing the gate capacitance and drive current. For its 45-nm technology, Intel deposits the hafnium-based dielectric using an approach that deposits one atomic layer at a time.
Known as Atomic Layer Deposition (ALD), this process provides exquisite control of the dielectric thickness. Intel claims the overall cost of its 45-nm process is in line with previous-generation process-based manufacturing trends.
The "secret sauce" that makes these HK+MG transistors unique is the combination of the hafnium's thickness, the type of hafnium, and the metals used for the gate electrodes. The combination determines the net benefits gained and dictates the reduction in leakage currents (gate and source to drain), the transistor density increase, and the switching speed increase.
None of the companies have disclosed any part of their respective combinations other than to point out that hafnium is being used to instantiate the high-k part of the equation. The only apparent certainty is Intel's feeling that it has the best "secret sauce," which will at least help keep the company one generation ahead of its closest rival and may even increase that gap.
Need More Information?