Physics and process drive etch performance at 45nm
03/01/2005
Next-generation scaling is translating into more exacting specifications for many processes in semiconductor manufacture. For dielectric etch, new materials, thinner layers, complex stacks of dissimilar films, and tighter tolerances are driving changes in applying physics to etch reactor operation and in process conditions to meet stricter demands. Effectively addressing these issues involves appropriate use of high-frequency source power, fully decoupled plasma-density creation and ion energy distribution, and ion-energy distribution tuning. By combining these elements with species distribution tuning, 45nm performance standards can be fulfilled for applications ranging from complex dual-damascene schemes using ultralow-k ( k<2.5) films to high aspect-ratio applications, such as contact and container etch.
Plasma etch requires the right combination of spatial power deposition (to create optimal plasma density and uniformity that, in turn, creates target etch rate and uniformity across the wafer); electron energy distribution (for etch selectivity with respect to photoresist);and ion energy distribution [for desired profiles and critical dimensions (CD)].
To maximize productivity - especially in multiprocess sequences, such as dual damascene, for which one-chamber processing is the ideal - two other factors also are vital. The first is optimizing power deposition in the plasma sheath on the wafer vs. other chamber surfaces, which promotes prolonged process stability and extends the lifetime of consumable parts, thereby lowering particle counts. The second factor is being able to generate a range of plasma densities appropriate for different purposes - etching a wafer, removing photoresist and residues, and cleaning a chamber.
Controlling plasma physics
For etching to occur, the reactor must create appropriately dense plasma. Although both source and bias powers create density, the bias frequency can limit the power available for density creation. Its primary effect is to increase the average ion energy delivered to the wafer. Isolating density creation to a high-frequency source and ion energy creation to a low-frequency bias offers several benefits.A high-frequency source can generate denser plasmas, even at relatively low power levels. Wafer etch rates are higher, although energy delivered to ions crossing the sheath on the source is significantly lower at high frequency than at radio (bias) frequency. Consequently, the source can simultaneously increase wafer etch rates while minimizing erosion of the source’s launching surface.
Lower energy bombardment is “kinder” to the wafer and the chamber, reducing the potential for damage (e.g., faceting) and sputtered buildup on chamber surfaces. Given that ultralow-k materials are much more porous and less dense than silicon dioxide, “softer” etch action better preserves their physical integrity. The same holds true for next-generation resists that are prone to roughening by ion bombardment and the consequent line-edge roughness and distortion in pattern transfer.
Separate, tunable source and bias power widens the etch process operating window, improving the capability to process successive layers of dissimilar films by switching from gentle to aggressive action as needed. For example, one needs high-source/low-bias plasma for organic etch, low-pressure/medium-density plasma for oxide etch, and high-source/low-bias plasma for chamber cleaning. Dual-damascene etch poses the most challenging test of flexibility given the different film combinations being implemented to overcome resist poisoning. Furthermore, the separately controlled high-frequency source overcomes the difficulty of obtaining the requisite density/energy mix at low pressures for etching and ashing ultralow-k films without damage and makes possible operating pressures from <10mtorr to the torr level.
Etch profiles depend not only on ion energy, but also on the distribution of energy. Some low-frequency bias systems (~10-15MHz) produce average ion energy of 1000V, with a spread of 900-1100V. Others, with bias frequencies of 0.5-2MHz, have similar plasma densities and average ion energy of 1000V, but with an energy spread of 200-1700V.
A combination of very low and low bias frequencies can produce a continuum of adjustment in the range of ion energy. Actual energy levels at the wafer have a significant effect on etch performance: Low energy levels lead to profile bowing, while undesirably high energies could result in an overly aggressive etch. Correctly balancing the bias frequency that produces a narrow range of energies (10-15MHz) with one that produces a wider range (2MHz) makes it possible to create a desired range of actual energies on the wafer without affecting the average (and without necessarily increasing plasma density). This is especially significant in obtaining the desired bottom CD, profile, and selectivity in 45nm high aspect-ratio applications (Fig. 1).
 Figure 1. Tuning the ratio of low and lower bias frequencies enhances profile control by concurrently optimizing etch rate and selectivity. |
While electrons do not control oxide etch, they can determine which molecules are dissociated. The dissociation, in turn, can determine what is etched - oxide or photoresist mask. Although the responsible mechanism is the subject of speculation, it appears that if the plasma is sufficiently dense, electron collisions convey sufficient energy from the higher energy “tail” of the energy distribution to bulk electron energy for dissociation of reactants to occur and film to be etched. If the density is low (1×1011cm-3 to 5×1011cm-3), however, the requisite energy transfer cannot occur and film selectivity is maintained.
If a multistep sequence (e.g., dual-damascene etch) is to be performed in a single chamber, high and low electron energies will be required for different etch steps. The source must be capable of operating over a range of power levels to generate both lower and higher density/dissociation. Low dissociation is needed for high-selectivity etching to avoid damaging 193nm-compatible resists and to stop on ultrathin barrier layers; high dissociation is required for ashing resist without damaging ultralow-k dielectrics and for cleaning the chamber.
To counteract the greater etch-rate nonuniformity that accom-panies higher etch rates, one must control plasma distribution in the chamber and ion flux distribution to the wafer. The key is to modify ion density distribution without increasing ion energy. This can be achieved by tuning charged species independent of plasma density and ion energy creation (Fig. 2), which also promotes the greater etch directionality required for smaller features, avoiding the low pressure/low ion density/high ion energy dilemma otherwise encountered.
As charged-species distribution influences etch depth uniformity, neutral species distribution affects process uniformity (Fig. 2). Gas flow velocity and pressure are the primary forces determining performance. Tuning them, independent of other process knobs, controls polymer deposition for desired selectivities and CD uniformity (<5nm repeatability within wafer, between wafers, and between chambers for 45nm).
Process complexities
The foregoing physics mechanisms mitigate some of the increasingly challenging conditions facing process chemistries. They create the wider operating window required for processing advanced dielectrics and enable processes to fulfill competing demands for polymer management in selectivity, etch profile/depth, and passivation as patterns are transferred multiple times through alternating layers of organic and inorganic films. The porosity of ultralow-k films is particularly challenging, as passivation must completely seal any open pores exposed during etch and create the desired profile to avoid defective device operation.
These performance demands can be more readily satisfied in a process environment that offers an ultrawide pressure range (<10mtorr to the torr level); source and bias powers that can range independently from 0-2000W and 0-5000W, respectively; high gas flows; and numerous (>12) independent process gas chemistries, especially in dual damascene. Such an environment also enables the process chamber to accommodate etching multilayer resist or metal hard-mask integration schemes being used to overcome resist poisoning, the greatest threat to etch performance at deep sub-90nm nodes.
Maintaining the low-k dielectric characteristic through the resist strip also is vital. Obtaining a clean, resist- and polymer-free surface without observable damage to the low-k dielectric requires an ultralow-pressure resist-stripping process regime and sufficiently dense plasma. Sustaining the required density at ultralow pressures requires the source to have a higher plasma-density creation capacity overall. In dual damascene, implementation of the requisite pressure/density conditions has proven beneficial in obtaining completely straight sidewalls in low-k films.
Mixing organic and inorganic films in next-generation film stacks requires multistep processing sequences that often involve incompatible chemistries. Correctly managing etch-chemistry transitions from one film type to another is critical to eliminate chemistry residues and memory effects. A continuously sustained, stable plasma is required throughout these etch/transition sequences to allow for rapid switching among differing source and bias conditions without the need to restrike.
 Figure 4. Preliminary parametric yield on 100nm via chains [a) 10mil and b) 1.3mil] shows yield >95%. |
Effective chamber cleaning is necessary to minimize wafer handling during complex organic and inorganic etches. Using a high-dissociation source power regime makes it possible to perform all-in-one dual-damascene processing and to sustain 45nm repeatability and particle count performance for this and other applications on a production scale.
Results
Figure 3 illustrates 45nm-compatible performance on several dual-damascene integration schemes achieved by applying the physics controls and process conditions cited previously. All were processed in single pump-down, single-chamber operation. The cleanliness of process behavior in a simulated production environment validates the feasibility of all-in-one dual-damascene processing in the fab. Preliminary electrical data on 100nm via chains from two-level copper structures with a multilayer mask dual-damascene integration scheme show parametric yield exceeding 95% (Fig. 4), effectively demonstrating the capability of fulfilling deep sub-90nm requirements in the completed structure.
Conclusion
Refined controls on etch physics create an operating window that accommodates the wider range of process conditions required for 45nm applications involving advanced dielectrics, thinner films, higher aspect ratios, multilayer masks, and other processing complexities. At the same time, these improvements achieve productivity gains by way of ultralow particle counts, and lot-to-lot uniformity and repeatability over hundreds of RF hours and thousands of wafers.
Robin Cheung received his MS in electrical engineering from the U. of Illinois and is director of technology in the etch products business group at Applied Materials Inc., M/S 81334, PO Box 58039, Santa Clara, CA 95054; e-mail [email protected].
Dan Hoffman received his PhD in plasma physics from the U. of Wisconsin-Madison and is senior director of engineering in the etch products business group at Applied Materials.