45 to 32 nm: Another Evolutionary Transition

Like the 45 nm generation, the 32 nm node will take advantage of the performance tricks of the prior technology generation, including immersion lithography, multiple stressors for mobility enhancement and porous low-k materials in the interconnect. The biggest change at 32 nm will be the introduction of high-k dielectrics and metal gates into production, though different approaches are likely to be taken. Other highlights are copper plugs and 3-D integration. This article will discuss these exceptions to the rule.

Lithography at 32 nm

Although there is no consensus on the chosen lithography technique for 32 nm, it seems clear that most parties are banking on the extension of ArF immersion lithography or double patterning, since extreme ultraviolet (EUV, 13 nm) is unlikely to be ready in time. The most aggressive geometry shrinks are taking place in the NAND flash segment, followed closely by DRAM and then high-performance logic devices (Figs. 1 and 2). Flash devices are characterized by the densest pitches, so the strongest resolution enhancement techniques (RETs) are typically used. In some ways, flash manufacturers have become the drivers of advanced lithography techniques, since they will be the first to incorporate them.

Most companies will introduce immersion lithography at the 45 nm node. As this issue was going to print, AMD (Sunnyvale, Calif.) and IBM (Yorktown Heights, N.Y.) announced that they had successfully completed their development work on the first-generation immersion lithography process using water, and that the yields on the new immersion tool were equivalent to those attained using dry ArF lithography. This is very significant, since it is early evidence that immersion lithography is likely to work in production and, just as importantly, will be delivered in time for 45 nm. According to Nick Kepler, vice president of logic technology development at AMD, much of the success can be attributed to its close R&D alliance with IBM, work performed with ASML (Veldhoven, Netherlands), and access to one of the first immersion tools at Albany Nanotech (Albany, N.Y.). Paul Agnello, project manager for 45 nm CMOS device and integration at IBM, stated that the most significant defects — air bubbles and wafer drying spots — had been taken care of. This advancement is part of the reason AMD will be able to ramp to 45 nm six months ahead of schedule, according to Kepler.

1. Because of accelerations in cell-size scaling, NAND flash has surpassed DRAM as the most aggressive CD.

The rapid progress in immersion lithography means that companies will make every effort to extend its use to the next generation. With water and the latest lenses, a numerical aperture (NA) of 1.35 can be attained. The lithography options at 32 nm involve immersion lithography and EUV (Table 1). However, since the usable range of k₁ in manufacturing starts at 0.25, water immersion will not meet all of the critical-level needs without resorting to pitch relaxation or double patterning. The next logical step, a change to next-generation immersion fluids (1.65 refractive index, [RI]) and higher final lens RI (1.9 from the current 1.57), will enable minimum resolutions of 39 and 35 nm, respectively (k₁=0.28). The latter would provide an NA of 1.55. However, this infrastructure is not likely to be in place in time to pattern 32 nm half-pitch features, so double patterning will become the bridge technology.¹ The key challenges here are the overlay requirement and process integration (see “Lith Experts Explore Options for 32 nm Half-Pitch ”).

2. The different imaging requirements of leading-edge devices. (Source: ASML)

Franklin Kalk, chief technology officer of Toppan Photomasks (Round Rock, Texas), said that some of the integration work will be a challenge for immersion, such as refreshing and recycling the water, which will become especially critical for much more costly next-generation fluids. He also said that problems, such as haze defects (molecular contamination on the mask surface), will be just as big a problem for immersion ArF as for dry.

These schemes could compete with EUV lithography at the 32 nm node, but will more likely face-off against EUV at the 22 nm node. The challenges for EUV include source power, optics lifetime, resist infrastructure and mask infrastructure.

Strained silicon

Last year at this time, we showed how various stressor techniques will be used at the 45 nm node to improve the mobility of charge carriers in the channel (electrons in nFETs and holes in pFETs).² Compressive strain is induced in pFETs using epitaxially grown silicon germanium (SiGe) in the source/drain (S/D) regions and/or a compressively stressed nitride layer over the gate. A technique called stress memorization uses poly implants and a stress transfer film that is annealed and removed, leaving a memorized strain. Tensile strain in nFETs is accomplished using increasingly strained nitride caps. Since then, although biaxial, global strain approaches on bulk substrates have not gotten much attention, the approach of strained silicon on insulator (SOI) appears to be gaining significant momentum.

3. A buried SiGe island is very effective at inducing uniaxial tensile strain in the nFET channel for a 15% improvement in drive current. The TEM shows the device following silicon regrowth in the source/drain. (Source: IBM)

Of course, once the drive current is achieved by strained silicon and other means, it is up to the low-resistivity junctions to carry this performance out through the interconnects. Junction engineering and optimization of silicides, even new materials, become critical as more strain techniques are employed.

High-k/metal gates

There are essentially three approaches to metal gate implementation: metal-inserted poly stack, fully silicided gate (FUSI) and dual-metal gate by replacement gate (Table 2). Companies are likely to take different approaches based on their product line mix (high performance vs. low power and low standby power), manufacturing expertise and roadmap for implementation. Low operating power and low standby power are strongly driven by cost, while it is possible that high-performance devices could use more exotic metal solutions.

At an Applied Materials' event at the International Electronics Device Manufacuturers Conference (IEDM) in December, Farhad Moghadam, senior vice president of Applied Materials (Santa Clara, Calif.), referring to the dual-metal gate option, stated: “The nFET metal gate has been proven from a materials standpoint, but there is no convergence on the pFET metal.”

Though there have been many papers published, very few companies have gone public with their plans regarding high-k/metal gate introduction into manufacturing. “At 32 nm, we have a clear intent to get there. We are currently moving into mainstream development,” said Suresh Venkatesan, director of Austin Silicon Technology Solutions at Freescale Semiconductor. Venkatesan admits that while identifying the best pFET metal has been a problem, there is not an inherent problem with the pFET itself. He says that through the use of a buried SiGe channel, strained SOI substrate and other strain techniques, good pFET performance and reliability can be attained without a metal gate.

FUSI approaches have reached the implementation stage for some companies. According to Serge Biesemans, director, CMOS Device Technology Group, IMEC (Leuven, Belgium), the keys to successfully implementing FUSI lie in understanding and optimizing the silicidation processes. At IEDM, IMEC demonstrated the manufacturability of dual FUSI gates, meeting the International Technology Roadmap for Semiconductors (ITRS) requirements for low power at 45 nm.⁴

One dual-metal gate option presented by Sony (Kanagawa, Japan) at IEDM⁵ used a damascene W/TiN/HfO₂ gate stack for the nFET device and a W/TiN/HfSi_x/HfO₂ stack for the pFET to attain drive currents of 1050 and 710 µA/µm at 1 V V_dd. Further integration of the pFET on (110) orientation substrate enabled even higher drive current.

Samsung (Kyunggi-Do, Korea) engineers presented an approach to metal-inserted poly stack at IEDM.⁶ In this process, after gate dielectric deposition, TaN electrode was deposited by advanced vapor deposition on HfSiON or HfO₂, followed by poly deposition. Drive currents of 620 and 230 µA/µm were attained for the nFET and pFET, respectively, with no added strain techniques. The study determined that low and symmetrical threshold voltages (V_t) could be attained without counter-channel doping, which complicates many dual-metal gate approaches. Part of the advantage of this process is the simplicity (no replacement gate and the use of existing materials). However, the new deposition technology had to be applied to attain the TaN with proper characteristics, and TaN thickness is critical because it has a strong effect on electrical parameters, including gate leakage, V_t and mobility characteristics.

On the fence with finFETs

As gate length approaches ~30 nm, the gate loses control of the charge in the channel. Two solutions appear at hand: ultrathin-body SOI and multi-gate FETs. Though several industry experts are opposed to multi-gate approaches, working circuits have been demonstrated at the 65 nm node.

“One of the questions we have is 'Is the finFET more variable than a planar device?' Once we determine the answer to that question, we can take the next step in evaluating this architecture,” Venkatesan said. Ghavam Shahidi, IBM fellow, has been outspoken on the topic of finFETs, stating that if 3-D transistors were going to be used, they would have been used by now and, reiterating his stance at the Applied Materials' event at this year's IEDM, said that finFETs will not be used at the 32 nm node.

What seems clear is that 3-D structures may not be applied throughout the industry at once. But there is strong motivation for going to finFET-type structures for certain applications, especially for analog/mixed-signal devices, where gain and linearity are critical performance metrics.

Multi-gate devices like finFETs have superior electrostatic integrity and control over short-channel effects. Essentially, by containing or fully surrounding the channel with the gate, the electric field is nearly uniform throughout the channel. As a result, the transistor behaves like a fully depleted device. Some of the advantages to the design are the complete lack of well implants, no latch-up, and reduced parasitic capacitance and resistance problems associated with shrinking space between transistors.

Infineon Technologies (Munich, Germany) recently demonstrated finFET circuit-level results (a 3000-transistor chip) with multi-gate finFET transistors fabricated in 65 nm SOI.⁷ Using a combination of metal gate and SiON dielectric, the fin was completely undoped. Most impressive was that there was nearly an order of magnitude (90%) reduction in leakage current with finFETs relative to the planar devices. This test demonstrated that potential roadblocks in finFETs — fin height, width control and process complexity — are surmountable. For Infineon, finFETs will have the potential to be useful for wireless RF devices at up to ~10 MHz. An Infineon spokesman has said that multi-gate devices will probably be necessary at 32 nm, and perhaps 45 nm.⁷

Another company that is actively pursing 3-D structures is Intel (Santa Clara, Calif.). At IEDM, they demonstrated a floating body cell (FBC) with independently controlled double gates that are self-aligned to the S/D for high-density embedded memory.⁸ The design is attractive because it features a small cell size and no storage capacitor (since charge is stored in the SOI body). This FBC is potentially easier to integrate than embedded DRAM.⁹ Intel noted that very aggressive scaling of body thickness, gate length and back-gate oxide thickness are required to simultaneously achieve high-speed, low-voltage and good-retention characteristics, although the design eases the scaling restraints of other FBC devices.

Intel fellow Tahir Ghani stated that existing strain techniques, in principle, should be able to be applied to finFET transistors. He noted, however, that they may be applied in different ways than they have been for planar devices.

Interconnects

The good news on the chip interconnect side is that there will be few process and architectural changes at the 32 nm node. The biggest changes occur with respect to the low-k dielectric in logic devices and the transition to copper interconnects for memories.

At the 45 nm node, leading device makers implemented the first porous low-k chemical vapor deposition (CVD) films (k~2.6) at the critical levels, which represented an enormous integration and engineering challenge for the fabs.¹⁰ For example, the low-power platform presented by STMicroelectronics (Geneva, Switzerland) and Freescale Semiconductor (Austin, Texas) uses ultralow-k (k~2.5) at M1-M7 levels, with TEOS at M8 and M9. In contrast, IBM's 45 nm interconnect stack uses a porous low-k selectively only at the switching-delay (RC)-dominated 2X levels (M4-M6), with low-k (k=3.0) at the 1X and 4X levels (M1-M3 and M7-M8) and fluorinated TEOS at the M9 and M10 levels. Likely extensions of these 45 nm stacks to 32 nm will involve enhancement of the barrier technology (making barriers thinner, yet just as reliable) and eliminating any assist layers, such as chemical mechanical planarization (CMP) caps and gradual improvements in the low-k. “As we move to 32, there are chances to optimize UV cure and provide more efficient removal of the porogen in the film,” said Tim Archer, senior vice president of the dielectrics group at Novellus Systems (San Jose).

4. The interconnect of a flash device shows on-pitch Metal 1 with high-aspect-ratio vias above, which press the capabilities of barrier/seed coverage and copper fill. (Source: Micron)

In terms of barrier and seed technology, barrier continuity is an ongoing issue in high-aspect-ratio features. Some memory manufacturers are looking at simple alloys of copper involving aluminum or magnesium (which migrate and become the barrier on anneal). Ruthenium-based oxides involve a more complex approach, but are still being investigated. Metal cap layers (electroless CoWP/CVD tungsten) face process margin and integration challenges. From a CMP standpoint, improvements are needed to polish tungsten caps and remove residual ruthenium seed.

Copper contacts

One of the significant changes in multi-level interconnects that may be implemented at the 32 nm node is the replacement of tungsten contacts with copper. Today, companies typically use titanium for improved contact, a sputtered TiN barrier followed by a tungsten nucleation layer and tungsten fill (CVD films). Copper plugs, with their lower contact resistance, alleviate the rapid rise in resistance as tungsten contacts scale below ~70 nm in diameter. Contact resistance impacts both RC and power consumption. As contacts scale, copper electroplating also provides an answer to the keyhole defects associated with tungsten deposition.

Of course, copper contacts must be properly contained within a barrier film, and not allowed to diffuse into the silicon and poison the transistors. According to engineers at IMEC, achieving adequate yield and reliability with a copper contact process requires an optimized barrier.¹¹ In this work, poor barrier quality was related to the formation of a copper silicide found in the S/D regions and around the gate. Poor barrier quality gave rise to yield loss in junctions and gate dielectrics and reduced time-to-breakdown.

A thin barrier is desired for lower contact resistance, while a thicker barrier ensures higher reliability. Therefore, a trade-off is required in manufacturing. Barrier resputtering, performed to thin the barrier at the bottom of the contact and improve sidewall coverage, can have an adverse effect on reliability.

3-D integration

In a keynote address at IEDM, Chang-Gyu Hwang, president and CEO of Samsung Electronics' semiconductor business, said that 3-D integration will drive the next stage of unprecedented growth in the semiconductor industry.¹² “Unlike the paradigm shift from the personal computer to mobile and digital consumer applications, the introduction of massive-scale fusion technology — which represents the organic convergence of information technology, biotechnology and nanotechnology — will bring together a wide range of technology-related professions as the foundation for a new technology frontier,” Hwang said. “Fusion will be enabled by 3-D technology integration. The core element needed to usher in the new age will be a complex integration of different types of devices, such as memory, logic, sensor, processor and software, together with new materials and advanced die-stacking technologies, all based on 3-D silicon technology.”

5. SEM of a 3-D via chain with 10,000 vias/mm2 density after etching the silicon in the top die. (Source: IMEC)

In this process, the copper vias are processed using copper damascene just after the contact layer and before the back-end-of-line (BEOL) interconnect stack. Vias are opened at the wafer backside by aggressively thinning the wafer down to 10-20 µm by grinding, CMP and etch. The wafer is diced by standard processes, and the dice are attached and electrically interconnected to a copper/dielectric landing substrate using thermo-compression bonding. This demonstration is important because of the density and interconnect sizes achieved by the potential extension to wafer-scale stacking.