Predicting the market in low-k dielectrics
04/01/2000
The path hasn't been easy, but there is little doubt that the market for low-k dielectrics is poised for take-off. Given low-k's embryonic nature, however, forecasting is subject to uncertainty. This article predicts the low-k market outlook for the next five years; reviews the recent history of low-k dielectrics; and makes a case for basing its forecasting technique on expected counts of dielectric film depositions and cost-of-ownership for various low-k alternatives.
The market
The semiconductor industry has actively investigated low-k dielectrics since the mid-1990s. It is only recently, however, that the need for these materials has reached a critical stage. Between 1999 and 2004, consumption of low-k materials for interlayer dielectric (ILD) applications is anticipated to grow from about $4 million in annual sales to approximately $300 million. Equipment expenditures during this period will total nearly $700 million on a five-year amortized basis. The outlook for the dielectric and low-k markets is shown in the figure.
Several positive factors have spurred growth in the market. First, semiconductor production has grown: The global trend is approximately 10%/year in units produced. Recession has had a minimal effect. Despite the 1998 Asian recession and a downturn in market values for that year, chip production was actually flat, and rebounded nicely in 1999.
A second factor is an increasing number of interconnect levels, especially evident in logic devices. Passivation and stress-buffer dielectrics are growing at a pace that matches overall growth. Highly integrated chips, however, must add interconnect layers in order to keep within the range of current die sizes. Another major consideration is the faster-than-industry-norm growth rates of highly integrated chips. Finally, low-k dielectrics sometimes need hard dielectric caps, resulting in two dielectric depositions in places where, formerly, one layer sufficed.
Today's low-k lineup
|
Dielectrics market, $ millions.
With 130nm devices nearing commercialization, true low-k integration is becoming a reality. Four classes of materials are in position to take part in the near-term ILD market, all of which rate k values in the 2.3-2.8 range:
- Silsesquioxane, particularly hydrogen silsesquioxane (HSQ), is an organosilicon compound that has already seen significant application as a spin-on that provides good gap-fill. Its low-k characteristics have been a secondary consideration until now. A big drawback is the risk of silanol formation in vias; recent improvements have dealt aggressively with this issue.
- PAE is a pure organic containing no silicon that has undergone extensive integration tests and is expected to go commercial in early 2000. Dow Chemical appears to be in the lead with its SiLK grade. Contenders are AlliedSignal/Honeywell International with Flare and Schumacher with Velox.
- Carbon-doped silicon dioxide, or carbon-doped glass, is the lone CVD candidate in the field, assuming FSG is dropped from contention. Although intensive integration tests are under way, they are not as advanced as those done on PAE and HSQ. The outlook is bright, however, because the process is more familiar to US-based fabricators. This dielectric is at least 10 times harder than spin-on organic or organosilicon types. Precursor reactants include organosilane or similar materials. After CVD, these materials create a silica structure that is opened up just enough by minor amounts of carbon to lower k. Even though the structure isn't porous, the effect is much the same.
- Silicon carbide (SiC), deposited by CVD from organosilane precursors such as trimethyl silane, is a likely candidate for replacing silicon nitride stop layers. Its k value is not in the same league as the above-mentioned products, but it is a big improvement over silicon nitride. When used with low-k ILD, SiC ensures a real improvement in effective dielectric constant.
The lineup four years from now
To achieve k values below 2.0, however, some degree of porosity will be required. The one exception to this is nonporous polytetrafluoroethylene (PTFE), under development at W.L. Gore & Associates and known as Speedfilm. (DuPont is also working on a CVD version of PTFE.) Recognizing this, materials suppliers are working to make their near-term candidates as "extendible" as possible by injecting porosity into them. Nanoglass, one of the original low-k candidates, has always been naturally porous as a result of the manufacturing process, in which a silica sol is spun onto wafers. As it sets, an organic solvent in the sol is driven off, creating pores.
The makers of silsesquioxanes and PAE have to look at different methods. Developmental efforts are focused on surfactants or dendritic polymers that would take up space in the matrix prior to cure and disappear pyrolitically during cure. The hope is to get k values down to <1.8.
IBM has been active in the development of dendritic sacrificial polymers. BF Goodrich has a class of sacrificial polymers of its ownnorbornenesmaterials that failed to make it as low-k dielectrics because of a low-temperature ceiling, but which withstand the initial bake well enough to set the porous structure and disappear during final cure. One or both of these types of technologies are being tried by Dow Chemical, AlliedSignal/Honeywell International, Schumacher on PAE, and Dow Corning on HSQ.
Meanwhile, other porous silica candidates have appeared from Asahi Chemical, Schumacher, and Pacific Northwest Laboratories. These are formed differently from Nanoglass and their pore sizes are smaller2-4nm on average, compared to 6-7nm for Nanoglass. Another big concern is mechanical strength. So far, none of the porous dielectrics have shown a modulus greater than 3GPa. It should be at least twice that high, and it would be better if it were four or more times greater.
How to predict the winners
Given this slate of contenders, how is it possible to predict market share in the face of rapidly advancing technological change? We based our predictions on the following findings in our study: CVD and spin-on deposition processes will coexist, in part because of regional preferences, but spin-on will win out when porous dielectrics come into use; the dielectrics with the greatest near-term potential are carbon-doped glass, organic polymers (especially PAE), and silsesquioxanes; and the lessened need for gap-fill in copper damascene encourages more CVD processing, at least in the near term before k-value requirements drop below 2.2.
We started by examining the basic demand for semiconductors. How many dielectric film depositions would be needed to satisfy this demand? Such data needed to be segmented by device typelogic, memory, and various subsegments of these main groups. For example, logic was segmented by microprocessors, application-specific integrated circuits (ASICs), digital signal processors (DSPs), and even sub-subsegments of these; memory was segmented by type and size (e.g., 1 Gbit DRAM).
By focusing on the more advanced semiconductor devices, those having multiple interconnect levels at design rules of <180nm, the low-k dielectric market was isolated.
Next, we converted chip demand to numbers of wafers processed, employing assumptions for die size and yield. Then, predictions for the count of dielectric layers, by type of layer (premetal, interconnect, hard mask, etch stop, passivation), were made for each market segment. The analysis also took into account the integration modeetch back or dual damascene. Some of these layers, particularly interconnect and hard mask, are candidates for low-k treatment. This analysis segmented the market for these layers by the design rules in effect for each of the device categories and, hence, the k-value range required of dielectrics used. The result of all this was a large set of film deposition counts by layer type and k-value requirements.
We then allocated the different types of dielectrics available to this market and estimated their total market value by consulting technical literature and the opinions of fabricators. Cost-of-ownership, which helped rank-order the alternatives and estimate market values, was examined in terms of capital and materials costs and stated as total cost/wafer film-layer. This process translated the deposition count into dollars of dielectric precursors and equipment.
Conclusion
Predicting the development of the low-k market is risky and requires the use of tools like Moore's Law, the ITRS, and market reports on the outlook for chip production. The market will ultimately settle on something that provides a k-value of <1.6 that should be the limit of technology (unless metal structures surrounded by air become possible). We believe that this dielectric will be porous and contain silicates for strength.
In the meantime, there are numerous avenues that fabricators might take while they can still manage with higher dielectric strengths; a material's extendibility will play a big role in choosing which to pursue. Over the next few years, the choice of best dielectric will be wide open and will need frequent review.
Michael Corbett is business manager of advanced materials and John C. Davis is a senior associate at Kline & Company Inc., Overlook at Great Notch, 150 Clove Rd., Little Falls, NJ 07424; ph 973/435-3457, e-mail [email protected].
Michael Corbett, John C. Davis, Kline & Company Inc., Little Falls, New Jersey
For more information on the source of this article, Kline & Company Inc.'s report The Global Outlook for Dielectric Materials in Semiconductor Devices, 1999-2004, contact Michael Corbett at the address above.