Solid State Technology asked executives to comment on the advantages and disadvantages of ALD vs. CVD techniques for gate dielectrics.
CVD and even PVD are still on the table
Luigi Colombo, TI Fellow, director of high-k development, Texas Instruments Inc., Dallas, Texas
The replacement of SiON gate dielectric is an onerous task. The industry will have to develop a new material and a suitable deposition technique for that material. The deposition technique selection will depend largely on the gate dielectric to be deposited and the ability to create a scalable, reliable, and producible gate dielectric.
A number of techniques have been used to deposit high-k gate dielectrics; however, the industry is focused on either atomic layer deposition (ALD) or chemical vapor deposition (CVD). ALD is relatively new to the semiconductor industry, but simple metal oxides such as Al2O3 for trench capacitors and metal nitrides for copper barriers have been demonstrated using this process. On the other hand, CVD is well established and has been in production for many years to deposit both dielectric and metallic compounds.
ALD offers excellent thickness control and conformality, but deposition temperatures tend to be much lower than CVD. Meanwhile, CVD offers excellent composition control and film homogeneity, and depositions are usually carried out at temperatures >400°C. In addition to CVD and ALD, there are still many reports on the deposition of high-k dielectrics using physical vapor deposition (PVD). While PVD has been used as an experimental technique and not seriously considered by integrated device manufacturers, the overall electrical results have been excellent.
A number of high-k gate dielectrics have been investigated and the industry is converging on either HfO2 or HfSiON. The choice of gate dielectric must precede the choice of deposition technique and ultimately the overall process flow. Many groups prefer HfO2 because of the expected higher dielectric constant; it is also a simple binary compound. Others prefer HfSiON because of its superior thermal and electrical stability. Recent reports also indicate that devices using HfSiON as the gate dielectric have a lower trap density, and silicates tend to have higher channel mobility than devices using HfO2. There are additional reports that a dielectric constant as high as 24 can be achieved for HfSiON with high-Hf and high-N films.
HfO2 has been deposited predominantly by ALD using hafnium chloride (HfCl4) and H2O as the oxidant. Other precursors, such as alkoxides and amides, have also been considered but are not as mature at this time.
The deposition temperature for HfO2 is about 300°C; impurity incorporation is a concern at this low temperature. Chlorine levels <1% have been reported; however, a potential issue with ALD HfO2 films formed by HfCl4 and H2O is the inclusion of OH that is typically associated with negative bias temperature instabilities. Another critical aspect of ALD HfO2 is that it requires a nucleation layer comprised of a few monolayers of SiOx to minimize the deposition incubation period. This may set a good interface between Si and the gate dielectric, but may also prevent ultimate scaling. HfO2 with metal gates has been scaled down to equivalent oxide thicknesses <1nm, but not with polysilicon electrodes.
CVD has been used to deposit HfSiO using Hf-t-butoxide and tetraethoxysilane (TEOS) or tetrakis(dimethylamido)silicon (TDMAS) and tetrakis(diethylamido)hafnium (TDEAH), among others. These precursors are liquids at room temperature and can be delivered to the deposition chamber through a vaporizer or bubbler. Films are generally deposited at temperatures >400°C, easing concerns about impurity incorporation. A major concern for the organic precursors mentioned previously is in-film carbon incorporation; however, carbon levels below the XPS detection limit exist in as-deposited CVD HfSiO films. Some CVD HfSiO processes do not require nucleation layers to initiate deposition, and the starting silicon surfaces are usually prepared by an HF-last process to minimize the presence of a lower dielectric-constant Si-O interfacial layer. A challenge of CVD HfSiO is that the films tend be rougher than HfO2 deposited by ALD. HfSiON films are formed by the nitridation of CVD HfSiO films by either plasma nitridation or ammonia anneal, and nitrogen levels as high as 25% have been achieved. HfSiON films formed by nitridation of CVD HfSiO have been used to fabricate scaled transistors with polysilicon electrodes <1.2nm.
PVD is not generally expected to be a robust gate-dielectric deposition technique. HfSiON, however, has been deposited directly using PVD, with results suggesting high-quality HfSiON films. Films deposited using the process have been scaled down to <1nm with polysilicon electrodes. Therefore, while PVD is not being seriously considered, it should not be discounted.
For more information, contact Luigi Colombo, Silicon Technology Development, Texas Instruments Inc., 12500 TI Boulevard, Dallas, TX 75243; e-mail email@example.com.
ALD is established, but challenges lie ahead
Tom Seidel, executive VP and CTO, Genus Inc., Sunnyvale, California
A high-altitude view of ALD shows the applications community in an expanding universe. In addition to semiconductor and magnetic-head manufacturing, ALD is being considered for MEMS fabrication and even applications outside microelectronics.
A large array of ALD thin films and processes with many oxides and nitrides have already been demonstrated, but a current focus [for Genus] is on Al2O3 and HfO2 — being used in the fabrication of dielectric films for DRAM capacitors, as high-k dielectrics for emerging metal gates, and as reader half-gap layers on magnetic heads. These all are enabling leading-edge applications of ALD.
Capacitors. ALD has been instrumental in extending deep-trench and cylinder-capacitor DRAM technology with Al2O3. Capabilities being developed include Al2O3-HfO2 alloys and nanolaminates. Work has involved thorough characterization of the TDMAH, TEMAH, and HfCl4 ALD chemistries with both H2O and O3 half-reactions. Deposition rates are comparable for all these precursors, and the ability of ALD nanolaminates to stand up to subsequent 450°C BEOL process temperatures has been examined: little or no interdiffusion of nanolaminates through a typical BEOL thermal budget is expected. Low carbon content has been verified in the resulting films deposited at low temperature. Work has shown that single-wafer TEMAH- and TDMAH-based ALD chemistries are capable of >95% step coverage with 40:1 trench aspect ratios.
Gates. For MOS gates, after a good investigative run at ZrO2, HfO2 is the core material of focus for high-k dielectrics in high-performance applications. Development work has demonstrated low, tightly distributed leakage for HfO2, and we are seeing <10Å equivalent oxide thickness with ~30Å ALD HfO2 as gate stacks are moving to alloys of HfAlON and HfSiON with metal electrodes. Engineers working on the development of HfO2 for high-k have learned how to control shifts in nMOS and pMOS threshold voltages by alloying other materials with HfO2. Additionally, the electron mobility of HfO2 can be favorably engineered using post-deposition annealing.
A remaining challenge with ALD HfO2 is coming up with thermally stable high vapor-pressure precursors or a delivery system that accommodates the existing precursors. If we had a hafnium precursor with the same vapor pressure and thermal stability characteristics as TMA for Al2O3, it would be a different world.
Magnetic heads. A critical requirement for thin-film heads is the dielectric for gap insulating layers surrounding active ferromagnetic films in magnetic head sensors. The dramatic rise in areal density in the disk-drive industry drove the conventional use of sputtered Al2O3 to its electrical and physical limits. (Areal density is the disk-drive metric defining information stored/unit area.) A paradigm shift came by adopting ALD Al2O3 films for reader half-gap layers. These films offer substantial improvements in dielectric breakdown, step coverage, and thermal conductivity compared to reactively sputtered Al2O3 films. These improved film properties have resulted in virtual elimination of wafer scrap from reader-to-shield shorts and improved reader thermal reliability in 80 GB/disc products.
For the future, engineers fabricating magnetic heads are looking at how ALD can be applied to next-generation "current-perpendicular-to-the-plane" (so-called "CPP") readers. Other novel applications for metal ALD films exist for both reader and writer structures in future transducer designs.
A forward-looking fundamental challenge. Today, ALD is at a threshold that dictates a greater understanding and extension of its chemistry and equipment-based chemical control. With trench capacitors, there is a challenge in the industry's road-mapped trend to further increase trench aspect ratio. Maintaining progress toward the 45nm technology node with its slated 93:1 trench aspect ratio will require ALD to deliver approximately 2× more precursor chemical/technology node to the wafer surface — totaling 23× more chemical into 45nm features in 2010. This must be achieved while keeping ALD wafer throughput at least constant, but preferably improved.
We see several pathways to solve this problem. One involves better understanding of ALD's saturating chemistries and control of half-reactions and the optimization of purge cycles. In one development pathway, we are finding it possible to maintain ALD-like processes with very small purge steps, thus significantly shortening cycle time and improving throughput. This process, called rapid ALD or RAD, has deposition rates of ~200Å/min, ~10× the throughput of classic, practiced ALD.
Another approach is to provide vacuum-engineered operating systems that have multilevel flow capability and, for example, a relatively higher flow during purge. Genus' first systems with such capability are now being shipped. The commercial implementation of this and other new ALD processes requires unprecedented co-development with subsystem component suppliers that are relied on for precursor delivery in ALD. The challenge of gas-switching valve reliability is keeping pace with ALD process innovations.
With current focus on challenges associated with capacitor applications, the results of the new work will likely have a broader application in enabling adoption of other thin-film solutions for ICs and other areas.
For more information, contact Tom Seidel, Genus Inc., 1139 Karstad Dr., Sunnyvale, CA 94089; ph 408/747-7140, e-mail firstname.lastname@example.org.