The market for low-k interlayer dielectrics
10/01/1999
In a recent market study of the use of low-dielectric-constant (low-k) materials for interlayer dielectrics (ILDs), we assessed the semiconductor industry's transition from the workhorse dielectric silicon dioxide (SiO2), through fluorinated silicon oxyfluoride (FSG) and hydrogen silsesquioxane (HSQ), to spin-on organic materials and inorganic systems deposited by chemical vapor deposition (CVD). Our findings are based on interviews with suppliers and potential suppliers of low-k materials, as well as with more than 20 semiconductor fabricators. For the purposes of the study, we defined low-k materials as having an intrinsic applied k of <3.0.
Implementation of low-k materials is developing more slowly than many suppliers originally expected. IC fabs' hesitation to use the materials and numerous technical challenges have resulted in the delay of low-k material adoption. The time for low-k dielectrics, however, is fast approaching. We estimate that the earliest products using low-k materials may arrive in late 1999 or early 2000, since many companies interviewed saw a need to make a commitment to a specific material by the third quarter of 1999. After this, low-k material consumption will gradually increase and start to show significant volume by 2001. By 2003, the global low-k material market should reach $150 million and expand rapidly.
 Figure 1. Forecast of short-term potential (columns) and actual (line) markets for low-k materials. |
The potential low-k dielectric market for leading-edge silicon will reach $338 million by 2003, but actual penetration may be half this figure. In the long term, low-k materials will be used not only for leading-edge applications, but also for memory and other logic devices. A forecast of the short-term potential and actual markets for low-k materials is shown in Fig. 1.
Semiconductor interconnect technology is gradually converging; aluminum and copper will co-exist in the next few years before the convergence is complete. After that, copper interconnection with low-k materials will be the mainstream technology for high-performance devices. For aluminum interconnect, spin-on organics would be the materials of choice, because inorganic systems deposited by CVD are only designed for copper damascene processing. Before the copper convergence is completed, the market for low-k materials will be driven by devices using aluminum metallization. We estimate that in 2001, 75% of low-k dielectrics will be used in aluminum interconnect devices; however, this amount is expected to decrease with time. When the copper-low-k era begins, both CVD and spin-on materials are expected to co-exist for some time. It is still too early to tell which technology will prevail, but it is clear that the whole industry will be characterized by uncertainty over the next few years. As one interviewee in our study stated, "My biggest problem right now is the confusion caused by the variety of material and process options that I face."
Low-k materials
Excluding the incumbent silicon dioxide and FSG and HSQ, there are three main contenders in the sub-3.0 low-k range: Dow Chemical's SiLK, AlliedSignal's Flare, and Applied Material's Black Diamond. Almost all semiconductor companies interviewed were working on some combination of these materials. The major candidates are shown in Fig. 2.
Applied Materials introduced Black Diamond last year in Hsinchu, Taiwan. Overall, its strongest selling point is that it is a nanoporous, inorganic, silica-based material applied by CVD processing. It is designed for copper damascene processing and is not for conventional aluminum gap fill technology. Black Diamond's k is 2.7; according to Applied Materials, this may be extended to the 2.4 to 2.2 range in the future by modifying the material's nanoporous structure. The company is using the material to address the extendibility and processing integration issues faced by the semiconductor industry.
AlliedSignal's Flare is a nonfluorinated, organic, spin-on dielectric material with a dielectric constant of 2.8. Flare is one of three products provided by AlliedSignal and is aimed at 0.18µm or finer generations. It is applied by a spin coater and reportedly has good thermal stability, up to 450°C, and superior crack resistance. In addition to Flare, AlliedSignal has two other materials that have a k <SiO2. One is Nanoglass with a k of 1.3 to 2.5, and the other is Accuspin T-23 LOSP with a k of 2.8 to 3.0. These three materials are targeted at high-performance IC devices with different dielectric constants. AlliedSignal has recently established a low-k integration facility in California to assist in low-k material implementation and processing integration.
Dow's SiLK is a nonfluorinated, highly aromatic, organic spin-on polymer with a reported isotropic k of 2.65 and very high temperature stability in excess of 450°C. SiLK is offered in addition to Dow's CYCLOTENE benzocyclobutene (BCB), which is the current material in production for GaAs ILD applications. SiLK formulations fit both copper/damascene and Al/W applications. Dow has recently built and started a new world-scale manufacturing facility dedicated to producing SiLK with the capacity to supply more than a dozen IC fabs in full production. Extendibility is claimed through the introduction of engineered porosity. Dow is the lead partner with IBM in a NIST-sponsored advanced technology program designed to develop and commercialize an optimized porous ILD system with <2.0 k.
The development of spin-on, organic, low-k materials formulated for both aluminum gap fill and copper damascene processing has been ongoing for several years. For example, Sematech, Leti, and IMEC have recently demonstrated integration with SiLK in copper damascene and aluminum/tungsten subtractive schemes in two-level structures.
Low-k material implementation
Everyone in the semiconductor industry recognizes that low-k material is required for the future; the only question is when it will be implemented. IBM and Motorola's 1997 announcement of their use of copper interconnects in high-speed SRAMs surprised many people who believed that low-k material would have been a better choice. The introduction of copper interconnects has complicated the low-k material development roadmap, but the use of lower-k materials to further improve device performance will be unavoidable.
 Figure 3. Convergence of semiconductor interconnect technology for high-speed and high-performance applications such as MPC, MCU, and ASICs. |
It was clear during our interviews that each semiconductor company has its own technology roadmap and low-k material. Before the convergence occurs at about the 0.13µm generation, the semiconductor industry will be fragmented in its choice of interconnect technology. Some companies such as Intel and TI will stay with aluminum interconnect and move forward in low-k materials technology. Other companies like IBM and Motorola are implementing copper interconnect first before they incorporate a lower-k material. The divergence of metallization choices combined with various low-k material options has made the technology development task more complicated. The ultimate goal, however, is consistent. It is agreed that low-k material will be implemented and combined with copper interconnect at about the 0.13µm generation by 2003. Figure 3 shows the convergence of semiconductor interconnect technology.
It is also generally agreed that low-k material will be implemented in devices that require higher speed or better signal integrity. Microprocessors, graphics processors, ASICs, complete silicon systems, and high-performance microcontrollers will be the first semiconductors using low-k materials. The specific choice of low-k material, however, is not clear. Most of the companies interviewed declined to reveal their precise low-k material choices. It is apparent, however, that the semiconductor industry will be slowly moving from traditional SiO2, through FSG or HSQ types of materials, to low-k materials. Adoption of FSG and HSQ materials with a k >3.0 has already happened in some devices such as SRAMs and microprocessors. Commercial availability or extended CVD equipment life explains the current adoption of FSG or HSQ materials in some devices. For low-k materials, however, SiLK, Flare, and Black Diamond are the leading choices.
US companies lead the rest of the world in implementing low-k materials. Several companies have already spent years in processing development and qualification, and are close to the product introduction stage. TSMC and UMC in Taiwan are also aggressively developing interconnect technologies, but are 6-12 months behind US leaders. Semiconductor fabricators in Japan and Europe are about 18 months behind in advanced interconnect technologies.
Acknowledgment
SiLK and CYCLOTENE benzocyclobutene are trademarks of Dow Chemical. Flare and Accuspin T-23 LOSP are trademarks of AlliedSignal. Black Diamond is a trademark of Applied Materials.
Shiuh-Kao Chiang, Charles L. Lassen, Prismark Partners LLC, Cold Spring Harbor, New York