Better HF Etch Uniformity with Single-Tank Approach

Isotropic etching of silicon dioxide with dilute hydrofluoric acid (DHF) is used repeatedly in the CMOS process flow for bulk etches, cleans and surface preparations. Bulk etches are used for controlled removal of many nanometers of SiO₂ for sacrificial-oxide strips, and the fabrication of dual-gate structures. DHF "cleans" are intended to remove 1 nm of native oxide or a few nanometers of bulk SiO₂, and thus expose and remove any metallics that are present in these oxides. DHF surface preparations are used to create a hydrogen-terminated silicon surface that is low in oxygen in preparation for gate growth, contact metalization or silicon epitaxy.

Historically, only bulk etches needed to be tightly controlled. The 2 to 3 nm of SiO₂ removed in cleaning or surface preparation steps has typically been negligible compared with device and film dimensions. Yet with dramatic device scaling, even a 3 nm clean can have a significant effect on device performance. As a result, the etch parameters of advanced films must be controlled as precisely in clean and surface preparation steps as it is in bulk etches.

In immersion systems, DHF etches have traditionally been performed in a conventional dual-tank approach by immersing a batch of wafers in a dedicated DHF bath and then moving the wet wafers to a separate rinse/dry tank. The temperature and concentration of this dedicated bath can be tightly controlled with closed-loop feedback techniques. Further, the repeated use of the bath for multiple batches of wafers can drastically reduce chemical usage compared with single-pass techniques for gross etches (>10 nm removal) and slow-etching materials such as silicon nitride. Careful control of bath concentration and temperature, etch and transfer times, and insertion and removal rates have resulted in uniform, repeatable etches.

The Magellan 300 STG immersion cleaning system offers chemical processing in a rinse/dry tank using SymFlow technology for high-uniformity, HF-last processing. The system architecture performs dilute HF etching in the same tank as the final rinse and dry, while maintaining a low-oxygen environment, keeping the wafers completely immersed until surface tension gradient (STG) drying is complete.

The process issues associated with the transfer from the etch bath to the rinse bath are of even greater concern. During transfer, hydrophobic silicon is very prone to particle deposition while passing through the DHF-air and air-water interfaces. Dissolved oxygen, in either the HF or the rinse water of pre-gate cleans, has also been shown to degrade device performance.

Ideally, the etch would be performed in a single tank with fresh DHF in the same rinse tank as the final rinse and dry — with the wafers maintained in a low oxygen environment, immersed in liquid until the drying process is complete. Such a conventional single-tank (CST) approach is implemented by immersing the wafers in a pre-mixed bath of DHF to soak for a pre-determined time. The etch is then terminated by an overflow rinse followed by a dry with a surface-active agent (typically IPA). The CST method is nearly ideal for the 3 nm etches used in HF cleans that remove the native oxide layer. The 2-3 min soak in a single-use bath of 200:1 HF does not significantly increase the cycle time in the chamber or use an excessive amount of HF, and it eliminates the extra tank and the liquid-air-liquid transition. However, the CST process has difficulty achieving the 3 ±0.2 nm etch uniformity required of advanced processes.

Most etch non-uniformity occurs during the transition of the wafers into and later out of the DHF, when local, time-dependent variations in etchant concentration cause local variations in etch rate. Incoming non-uniformities result from the finite insertion time of the wafers into the bath; the bottoms of the wafers are exposed to DHF ~1 sec before the tops of the wafers. This extra second of exposure to DHF results in a systematic 1% across-wafer variation for a 100 sec etch. Also, the DHF can be locally diluted by the incoming carryover layer of water from the previous rinse. The non-uniformity of wafers etched in 200:1 HF after an SC1/rinse or rinse-only step is typically 0.2 nm higher than that of incoming dry wafers.

After the transition in, etching of the wafers in DHF proceeds relatively uniformly while the wafers are completely immersed in a well-mixed bath. It is theoretically possible to uniformly terminate an etch by a rapid "infinite dilution" of the DHF with an overflow rinse of DI water, but only if the bath remains very well mixed. In currently available hardware, DI rinse water is introduced through arrays of small holes in sparger bars located at the bottom of the tank. The momentum of the water jets drives mixing within the bath. Unfortunately, the action of these jets typically does not keep the bath sufficiently well mixed, and substantial non-uniformities occur during CST rinsing.

We immersed full cassettes of 300 mm wafers into a CST system with a pre-mixed bath of 200:1 DHF and began the rinse process immediately. There was significant across-wafer non-uniformity with the CST transition-out rinse process with 80 and 40 L/min flows (Fig. 1a and b), but the higher flow produced more uniform etch rates. Data are shown for three wafers in slots 2, 26 and 51 of the 52-wafer cassette. The 80 and 40 L/min CST rinses removed an average of 1.05 and 0.55 nm of oxide, respectively. The total removal range across the wafers scaled almost linearly with the inverse of the rinse flow rate.

1. Across-wafer etch variation on three 300 mm wafers during the transition out from a 200:1 DHF to an 80 L/min rinse flow (a), and out of a 200:1 DHF to a 40 L/min rinse (b). Increased etch rate uniformity is possible with a single-tank, 200:1 DHF process with 40 L/min rinse flow (c).

We estimated oxide removal for a 120 L/min rinse flow to be 3 ±0.2 nm, which is required for advanced processes. Although it is possible to support DI flows of 120 L/min and higher, in manufacturing, the implementation creates difficulties. Piping and valving become so large as to resemble those used in the DI water plant. An intermittent demand of 120, 240 or even 360 L/min from multiple baths in a system can challenge the pressure stability of the local DI loop as well as the capacity of the waste system. Finally, the cost of sizing such large DI and waste systems and the ongoing cost of consumables is significant. Thus, a more practical solution is needed.

A review of the similarities in etch profiles seen in Figure 1a and b suggests another possibility for improved etch uniformity. While there is some difference in the etch patterns due to the difference in the flow dynamics created by the 40 and 80 L/min rinse flows, the overall trends match. For instance, points 18-28 show an area on the wafer with rapid rinsing and little etch, whereas points 29-35 represent an area that is rinsed more slowly and has a higher average etch during the process. Our tests showed that the etch profile of each slot in the cassette is consistent from run to run, particularly if the rinse flow rate is held constant.

It is possible to use symmetry to substantially cancel the fluid-dynamic effects that create this non-uniform etch profile, as was incorporated in the SymFlow design (Fig. 1c). We can invert the profile of the rinse process with the understanding that flowing DHF into a DI water bath creates a mirror-image concentration profile of flowing DI water into a bath of DHF. Areas that receive HF first during the injection of DHF are the same areas that receive DI water first at the beginning of the rinse. The process sequence to exploit this symmetry, compared schematically with CST (Fig. 2), is as follows:

1. Immerse the wafers in DI water.
2. Flow DI water to establish a stable fluid velocity profile in the bath.
3. Inject concentrated HF into the DI flow stream to create a flow of DHF into the bath.
4. Continue the flow of DHF as necessary to reach the desired etch target.
5. Stop the HF injection and allow the continuing DI flow to rinse the wafers.

This symmetrical rinse-etch-rinse process, SymFlow, produces a faster, more uniform etch profile (Fig. 1c). The non-uniformities of the transition into the DHF etch from DI are cancelled by the transition out of DHF into the DI rinse. At a total range of <0.3 nm, the uniformity of this SymFlow process at 40 L/min is comparable to that projected for the CST processes of Figure 1a and b with a 160 L/min rinse.

2. Etch rate non-uniformities in the CST approach result from transfer in and out of the DHF batch and insufficient mixing during rinsing. The SymFlow approach relies on symmetry between the HF infusion and the rinse to improve uniformity.

Figure 3 compares the total on-wafer etch range for etches of 200 mm wafers in a dedicated, recirculated HF bath with 300 mm data for CST etches with 80 L/min rinses and SymFlow etches with 40 L/min rinses. Even at half the liquid flow rate, the SymFlow etch is substantially more uniform than the CST etch and is comparable to 200 mm data from a dedicated DHF etch bath.

3. Average range (max-min) uniformity using a dual-tank, 200 mm system with recirculating HF is 2.4 Å (a); in a conventional single-tank, 300 mm system with 80 L/min rinse it is 4.0 Å (b); and it is 2.4 Å in the SymFlow 300 mm system — even with a rinse flow of 40 L/min (c).

We characterized the SymFlow system for carrier loading effects (Fig. 4). Typically, partial loads of product require the use of filler wafers to offset wafer position non-uniformity. However, filler wafers are expensive, require significant storage space and loading time, and can generate particulates after being processed hundreds of times. To eliminate this need, we measured the average etch rate change across-wafer in three tests in the SymFlow system — by running wafers in a fully loaded carrier (Fig. 4a), a test with end wafers (2 and 51) facing out without end cap wafers (1 and 52) (Fig. 4b), and with wafers in a half-loaded carrier (26/52 slots filled) (Fig. 4c). We detected no difference between the treatments.

4. Whether fully loaded (a), loaded with position 2 and 51 facing out without end wafers (b) or partially loaded (c), etch uniformity in the SymFlow system is maintained.

Process options

In many processes, wafers will require some cleaning before the DHF etch. For instance, ozonated DI water (DIO₃) is often used to remove light organics, followed by an in situ rinse, DHF etch, rinse and dry. If desired, DIO₃ also can be used after the etch process to regrow a passivating oxide on the wafers. In other cases, treatment in an ex situ bath is desired. Figure 5 shows the etch performance of a two-bath process, an SC1 treatment in a dedicated bath with recirculation and megasonic excitation followed by a rinse, etch, rinse, dry in the etch/ rinse/dry chamber.

5. Average etch range for a 5 min, SC1 step (1:1:50) at 55°C was 1.4 Å compared with 2.4 Å processed in the SC1 bath followed by SymFlow rinse, HF, rinse, STG dry.

Because the SymFlow process starts in DI only, the SC1 rinse can be performed in the etch/rinse/dry chamber and a dedicated SC1 rinse bath is not necessary. The dump rinsing function of the etch/ rinse/dry chamber can be used to greatly speed rinsing of the hydrophilic wafer surfaces after DIO₃ or SC1 treatments.

Unfortunately, dump rinse techniques often cause particle issues on wafers with hydrophobic regions. Therefore, overflow rinsing is generally preferred after a DHF etch that exposes hydrophobic silicon. Rinsing in a well-stirred system follows the conventional single-tank rinse (CSTR) theory, and the concentration drops relatively slowly as:

C = C₀exp(-Qt/V)

where C₀ is the initial concentration, Q is the DI flow rate, t is the rinse time and V is the volume of the bath. Each full bath of rinse water reduces the concentration by 63%. A tenfold concentration decrease requires 2.3 bath volumes. However, it is not necessary, or even desirable, to fully rinse the wafers (to 18 MΩ resistivity) after DHF etch. In fact, small quantities of HF in the final rinse are desirable to suppress the deposition of metallics from the water and to suppress the exchange of OH- for F- on the surface. Any residual traces of HF will evaporate with the water. The primary requirement is that the HF be dilute enough so as not to create etch non-uniformities during the surface-tension gradient (STG) drying process.

In an STG dry, the rinse water is slowly drained (at 2-3 mm/sec) from the closed rinse tank and replaced by a low concentration of IPA vapor in nitrogen. The SymFlow process etch uniformity is a function of rinse time. Figure 6 shows that, after rinsing for 2.3 min, the concentration in the bath dropped tenfold to 2000:1 with an effective thermal SiO₂ etch rate of 0.125 nm/min. This HF concentration would create a systematic 0.25 nm excess etch at the bottoms of the wafers during the 2 min STG drain step. Increasing the rinse time to 4.6 min reduces the HF concentration to 20,000:1 and the systematic error to 0.025 nm. Although the uniformity results in Figure 6 are near the noise level of routine ellipsometric measurements, there does appear to be some improvement in the uniformity between a 2 and 4 min rinse. Depending on the required etch uniformity, final rinses of 2-4 min are recommended. Increasing the flow rate of the rinse water can shorten the rinse time.

6. In the first 2.5 min of rinsing, the concentration in the bath dropped tenfold (to 2000:1) with an effective thermal SiO2 etch rate of 0.125 nm/min.

The primary drawback of the SymFlow process is HF usage. In a CST etch, 200 cm³ of 49% HF are required to create a 200:1 blend in a 40 L bath. Yet in a SymFlow etch, the DHF flows continuously for the entire etch period. A 3 nm etch typically requires a 4 min injection of 200:1 HF at 40 L/min for a total HF usage of 800 cm³ of 49% HF. However, the increased HF cost is more than offset by the lower DI water usage, reduced hardware costs (smaller facilities and one less bath) and, most significantly, by the increased device performance provided by the greatly improved etch uniformity.

If desired, the HF usage can be reduced by 50% through operating the etch at 40°C instead of at room temperature or by stopping all flow after the first 2 min of the etch cycle. By 2 min, the bath is well mixed with a concentration ~90% of that of the incoming DHF. As noted above, etching proceeds uniformly while the wafers soak in a well-mixed bath. Stopping the DHF flow and then later resuming the rinse flow has little impact on etch uniformity, but reduces HF usage by >50%.

In conclusion, the SymFlow process utilizes the symmetry of the transition into DHF and transition out to rinse water steps to greatly improve the uniformity of an etch performed in an etch-rinse-dry chamber. The process reduces tool complexity and footprint through elimination of the dedicated HF bath, and an additional dedicated rinse bath if SC1 or DIO₃ pre-treatments are used before the etch. Most important, SymFlow boosts device performance by providing greatly improved etch uniformity, while maintaining the wafers in an oxygen-free environment.