How to Minimize Resist Usage During Spin Coating

		At a Glance
	This paper presents a practical approach for minimizing the volume of photoresist dispensed during spin coating of silicon wafers. We have developed several reduced consumption processes for a wide range of resist viscosities (7-55 cps). This has resulted in a > =45% reduction in resist usage.

O ne of the most critical roles for spin coating systems is to properly cast a thin film of photoresist on the surface of a silicon wafer. This film must be extremely uniform in thickness(<6 Å, 1s), and wafer-to-wafer mean thickness control must be better than 30 Å total indicated range (TIR, max-min) over extended periods of time. In recent years, as the semiconductor industry has moved from moderately priced i-line resists to high-priced DUV-resist, the cost of chemical usage has driven production fabs to look closely at reducing the amount of resist required to achieve these goals.

We will first look at the market's perception of how much resist is required to coat a wafer and consider the economic engine that drives our industry (cost-of-ownership). Then we will review the fundamental physics of the mechanisms that control resist coatings and the impact of environmental, equipment and resist properties on this dynamic process. Utilizing this basic understanding, we will review a methodology for reducing resist consumption for any coating process and summarize some typical usages with common chemistries employed on SVG coating systems.

Market perceptions of resist usage

Many production fabs utilize 4-5 cc of resist for each 200 mm wafer. For less critical layers where i-line or g-line resists are employed, this is probably very cost-effective, since a wide process latitude and minimal spin defects will be obtained by using such large volumes of resist. However, for critical layers(<0.25 µm CDs) where DUV resists are typically employed, a significant savings can be realized by reducing this volume into the 1.5-2.5 cc range. In fact, many fabs have accomplished this goal, as shown in Figure 1.

Figure 2 illustrates the relationship between minimum resist usage and viscosity. In general, the amount of resist required to coat a wafer will decrease as viscosity decreases.

Fig. 1. Many fabs have accomplished the goal of using only 4-5 cc of resist for each 200 mm wafer.

Fig. 2. The amount of resist required to coat a wafer will decrease as viscosity decreases.

Fig. 3. Costs associated with photoresist use are expected to increase dramatically as the industry pushes to smaller geometries..

This has led to much confusion in the industry as process results are often quoted without specifying resist type. A high viscosity resist such as JSR300 (55 cps) requires 2.4 cc for good coatings, while a low viscosity resist such as TOK (8 cps) may perform well with as little as 1.3 cc. Concurrently, the lower viscosity resists also have a thinner target thickness.

Cost-of-ownership (COO) analysis

The goals of minimizing resist volume during the coating process are basically the same as any other materials utilized to manufacture microelectronic devices: reduce manufacturing costs through reduction in consumption and consequent waste disposal costs. One of the most important reasons to look at minimizing the usage of resist is the annual cost savings for consumption and safe disposal of the unused portion. Table 1 summarizes the resist cost per track on an annual basis for various dispense volumes; the conclusions are quite compelling. Figure 3 visualizes the escalating cost-per-wafer for DUV lithography producing sub-0.25 µm (250 nm) line-widths.

Fundamentals of spin coating process

Figure 4 shows four basic stages of a typical spin coating process, in which a stream of photoresist is dispensed onto the surface of a spinning wafer; these are deposition, spin-up, spin-off and solvent evaporation. The first three stages are somewhat overlapping and sequential in that dispensed photoresist outflows convectively, under the influence of centrifugal force, covering the entire surface. Solvent evaporation, on the other hand, continues throughout the process.

Table 1. Annual Cost of Resist Dispensed per Track
Calculated for a wafer throughput per year for one wafer track of (60 wafers/hr) (360 days/year) (0.90 track utilization) = 466,560 wafers/year
Resist material	Approx. cost/gallon	Cost/cc	1.0 cc/wafer	1.5 cc/wafer	2.0 cc/wafer	3.0 cc/wafer	4.0 cc/wafer	6.0 cc/wafer
SPR 510	$560	$0.148	$69,051	$103,576	$138,102	$207,153	$276,204	$414,305
Apex E	$1500	$0.396	$184,758	$277,136	$369,515	$554,273	$739,031	$1,108,547
DUV	$2000	$0.528	$246,344	$369,515	$492,687	$739,031	$985,375	$1,478,062
DUV	$3000	$0.792	$369,516	$554,273	$739,031	$1,108,546	$1,478,062	$2,217,093
DUV	$4000	$1.056	$492,687	$739,031	$985,375	$1,478,062	$1,970,749	$2,956,124
DUV	$5000	$1.320	$615,859	$923,788	$1,231,718	$1,847,577	$2,463,437	$3,695,155

During the spin-off stage, the resist film is thinned down to a thickness whose uniformity profile is primarily determined by two competing mechanisms (mass transfer and evaporation). According to this figure, convective outflow is the dominant thinning mechanism initially. However, increased viscosity because of relatively small but constant solvent evaporation causes convective outflow to decrease with time. A few seconds after the end of dispense, solvent evaporation becomes the dominant mechanism and determines final film thickness and uniformity. The thinning rate because of evaporation starts to decrease after about 5 sec from deposition because of lowered diffusivity of the remaining solvents within the semisolid photoresist film.

Fig. 4. The first three stages of the spin coating process are sequential in that dispensed photoresist outflows convectively under the influence of centrifugal force. Solvent evaporation continues throughout the process.

Successful implementation of a reduced photoresist consumption process requires a detailed understanding of these mechanisms. At the end of a coat process, the film still contains some solvent, and its thickness changes dynamically as it interacts with its local environment until it is delivered to the softbake module, where final target thickness, uniformity and solvent content are achieved. The coating is less sensitive to the environment following the soft-bake step.

This picture of two competing mechanisms starts changing as we lower the volume of resist dispensed on a silicon surface. This is shown in Figure 5, where the final film thickness is displayed as a function of initial wet thickness (volume) for three different spin speeds. ¹ Roughly speaking, the final film thickness is independent of the initial wet film thickness (dispense volume) of 60-1000 mm (-2.0-30 cc for a 200 mm wafer) for all three spin speeds. This is consistent with the two-step mechanism, which is valid for any conventional large dispense volume process. As the dispense volume is lowered even further, final film thickness starts becoming dependent on the dispense volume. The three curves in Figure 5 join where the initial wet film is so thin that evaporation dominates convective outflow immediately after the dispense.

Fig. 5. Final film thickness is generally independent of the initial wet film thickness, regardless of the spin speed.

This roughly corresponds to 10-30 µm wet film thickness (~0.3-1.0 cc dispense volume) for the viscosity used in this calculation; therefore, there is a paradigm shift from a two-step process (convective outflow and evaporation) toward a single step (evaporation) as the dispense volume of resist is reduced. It is this lack of an effective convective outflow, thus lack of 100% coverage, that ultimately limits the minimum dispense volume one can obtain. The imbalance between these two mechanisms is also a source of defect generation.

Resist minimization guidelines

The following is a summary of the guidelines for resist minimization based on the fundamentals of mass transfer discussed above:

• Process optimization needs to be done in the two-step regime. This can be done first by identifying the cut-off volume, below which there is no full coverage, followed by determining a minimum volume for which the two-step process is valid.
• The cut-off volume below which there is no full coverage is viscosity dependent. The lower the viscosity of the liquid polymer, the smaller the cut-off volume.
• Select only the first-order parameters that have an impact on convective outflow and solvent evaporation mechanisms for the statistical process optimization by means of design of experiments (DOEs). The first-order parameters are dispense spin speed, exhaust flow rate during dispense and drying steps, dispense rate, resist temperature, chill plate (CP) temperature before coat (wafer temperature) and ambient air temperature within the coater.
• Other significant process variables that are not included in the optimization process need to be controlled tightly at an established optimum value. These variables include ambient air humidity and airflow velocity.
• Tool setup-related parameters such as nozzle size, nozzle height, nozzle centering and various coater-related calibration parameters need to be set properly, similar to conventional high-dispense volume coaters.

Experimental methods

Results of extensive resist minimization studies are summarized in Table 2. A wide range of resist materials with different viscosities were studied. Investigations started with fundamental experiments of resist dispense dynamics using a color high-speed camera in order to gain a detailed understanding of the basics of spin coating. These first principle understandings were then combined with the statistical DOEs in order to develop optimized low consumption photoresist processes.

The resist dispense system tested contained a 5/32 in. (I.D.) nozzle and a IDI300 dispense pump. Macroscopic visual defects were checked under monochromatic light with an optical microscope. Microscopic defects were analyzed via optical scattering inspection (Surfscan). Temperatures were measured by calibrated Pt RTDs, and resist thickness profiles were measured by Therma-Wave and/or Prometrix film thickness measurement tools.

Table 2. Summary of Resist Minimization Results
Coater system	Resist type	Viscosity (cps)	Volume dispensed (cc)	Target thickness	Uniformity1 (Å)	Range 2 (Å)
200APS	TOK	7	1.3	8000	5.99	30.1
90S	SPR505	8.2	1.5	7600	4.18	25.7
90SE	SPR505	8.2	1.4	6000	6.0	25.5
90SE	SPR507	12.3	1.9	8400	5.15	21.7
90S	AZ7511	10.1	2.1	108,000	6.25	40.5
90SE	APR508	13.9	1.4	9950	7.6	34.0
200APS	SPR508	13.9	2.1	10,000	6.13	19.0
90SE	SPR510	18.6	2.1	12,000	6.64	26.7
200APS	JSR061	18.0	2.1	10,600	6.48	28.5
200APS	JSR300	55.0	2.4	32,000	30	85

The following experimental results outline first-order parameters that impact resist coating performance and detail their influence on resist minimization methodologies.

Environmental factors

Fig. 6. The dependence of final film thickness on resist temperature is complicated; increasing resist temperature decreases viscosity (which results in a thinner final film) but increases evaporation rate (which results in a thicker final film).

Resist (Tr), air (Ta) and chill plate (Tw) temperatures -- There are two competing mechanisms that cause complicated resist film thickness profile dependence on the Tr. Viscosity comes down with increasing resist temperature, causing more effective convective thinning of the film, which lowers the film thickness. Solvent vapor pressure (evaporation rate) goes up with increasing Tr; this causes an increase in the thickness. This is usually a localized effect and is more dominant near the vicinity of point-of-dispense. The magnitude of each effect depends strongly on the resist viscosity and solvent volatility. Both effects are clearly visible in the response curve shown in Figure 6. For a fixed CP temperature of 20.7°C, mean film thickness for a Tr = 21.7°C is about 350 Å lower than that of Tr = 20.7°C. This is consistent with the improved convective thinning discussed above. However, the in-creased thickness at the center for Tr = 21.7°C is consistent with the enhanced evaporation at the center. In general, for lower viscosity materials the second mechanism becomes dominant. In that case, resist temperature is effectively used as a film thickness uniformity control knob to compensate for enhanced evaporation rate at the wafer edge because of high linear velocity.

The discussion above also applies to CP temperatures in that it basically determines the wafer temperature before the dispense of photoresist. A CP temperature set similar to air temperature will minimize heat transfer between the wafer and ambient and will ensure that the wafer temperature before dispense is stable and repeatable, as long as there are no significant heat sources in the vicinity. A resist temperature set slightly higher (Tr = Tw + 0.7°C) is expected to spread resist better because of a slight decrease in viscosity. It is advisable to use a DOE to optimize these three temperatures together.

Relative air humidity -- Increased relative humidity suppresses the evaporation rate, causing the mean film thickness to decrease across the wafer. The effect on the film thickness uniformity is negligible. In these experiments, relative air humidity was fixed at 40% and tightly controlled to within ±1% of the set point for stable mean thickness control.

Fig. 7. Higher exhaust flow rates tend to increase linear flow velocity at the edge of the wafer, which can cause increased film thickness at the edge.

Fig. 8. The dependence of resist film thickness and uniformity for two different spin speeds is shown.

Spin exhaust flow -- Linear airflow velocity for a spinning wafer increases linearly with position along the wafer radius, making the edge of the wafer have the highest airflow velocity for any spin speed. Higher exhaust flow rates will tend to increase this velocity even further. This causes increased film thickness toward the edge of the wafer, because of the enhanced solvent evaporation rate (Fig. 7). The exhaust flow rate of 800 liters/min gives increased thickness at the edges, whereas 200 liters/min produces lower thickness near the edges.

Equipment variables

Dispense spin speed (DSS) -- The DSS determines the magnitude of the centrifugal force acting on the dispensed resist, which counters viscous forces. In addition, increased DSS is expected to increase solvent evaporation rate because of increased linear velocity causing an increase in the overall film thickness. Figure 8 shows the DSS dependence of resist film thickness and uniformity for 3000 rpm and 4000 rpm with a dispense rate (DR) of 0.8 cc/sec. Film thickness is ~100 Å higher for 4000 rpm, and resist profiles at the edges show an increased thickness because of enhanced solvent evaporation. DSS is an important variable that should be changed together with DR in order to minimize dispense volume.

Acceleration and post dispense delay -- Timing of the transition to the final spin-off stage is a crucial one that needs to be fast and repeatable for thickness control. It was established that for a viscosity range of 12-19 cps, the dependence of mean thickness on post dispense delay is about ±100 Å/sec and is expected to increase for higher viscosity resists such as JSR300. Mean thickness increases (+100 Å/sec) occur when the final drying spin speed is higher than the dispense speed and decreases when the drying speed is lower than the dispense speed. From our studies, 500 rpm/sec was found to be the optimum acceleration for the drying step.

Fig. 9. The dependence of average film thickness on dispense rate (for a fixed spin speed of 3000 rpm) is shown.

Dispense rate (DR) -- Figure 9 shows the dependence of average film thickness on DR (for a fixed DSS of 3000 rpm). A high DR (3.2 cc/sec) tends to give higher mean thickness. Low DRs (0.8 cc/sec) are influenced by solvent evaporation. In addition, a high DR tends to give lower thickness profiles at the point of impact (center). This may be due to the mechanical impact of the resist with the wafer and the subsequent bounce of the resist stream. It turns out that the dependence on the DR becomes much weaker as the resist viscosity decreases. Experiments using variable DRs are under way.

Dispense type -- Center dynamic dispense methods were employed in all of these experiments because of its robustness and repeatability.

Resist and surface variables

Solvent volatility -- Solvent volatility is an important variable in that it determines the solvent evaporation rate during and after a coat process. Up to 70% of JSR300 is an organic solvent (ethyl lactate) with a vapor pressure of 2 Torr at 20°C. The relatively low vapor pressure of ethyl lactate provides improved process latitude and robustness.

Resist surface energy -- The HMDS process lowers the total silicon surface energy from 50 to 25-30 dynes/cm. In order to improve spreading of the liquid resist, the surface energy of the resist formulation should match that of silylated-silicon. The surface energy of the ethyl lactate, the main ingredient of JSR300, is 32 dynes/cm.

Resist viscosity -- Resist viscosity is one of the most important variables for a resist minimization process. Resist viscosity is time-dependent in that it increases as a result of continuous solvent vaporation during the dispense and spin-off stages. In addition, the impact of the two competing mechanisms varies with viscosity. Every other variable being the same, convective outflow becomes more dominant for lower viscosity materials and evaporation is more dominant for high-viscosity materials.

Surface type -- The surface energies of films vary, thus changing their wetability by organic polymers such as photoresist. In addition, the topography of the surface brings added complications to the convective outflow mechanism because of the spatial variations on the underlying surface. This work was performed on bare silicon; however, the methodology described at the end of this paper has also been applied for various films with topography.

Methodology for reducing resist consumption

Now we will look at the whole picture and try to establish a methodology that any process engineer could use to reduce resist usage. Overall, the strategy is to establish a minimal dispense volume from a basic understanding of the principles outlined above, then optimize the process latitude by DOE studies. It is important not to get caught up in the game of only trying to minimize resist dispense volume and wind up with an unstable process.

The following procedure outlines the most significant variables to consider when trying to reduce resist usage from, for example, 4.5 cc to 2 cc. This six-step method will allow a process engineer to establish a coat recipe that will yield well-controlled thickness uniformities and the largest process latitudes.

Step 1: Preliminary starting process -- Set Tr = Ta = Tw = 21°C, RH = 40%, air velocity = 75 (on Semafab), EFR = 200 liters/min (dispense and drying), DR = 0.8 cc/sec, volume = 2.5 cc (dispense time = 3.125 sec) and DSS = 1200 rpm. Use center dynamic dispense recipe.

Step 2: Achieve 100% coverage, with no visual defects. Coat five wafers to make sure you have 100% coverage on all the wafers. If not, increase your dispense rpm by 100 rpm increments until you have full coverage. For viscosities of 10-20 cps you may need 1200-1600 rpm, whereas for 40-60 cps you may need 2500-3500 rpm.

Step 3: Estimate minimum volume for 100% coverage (Vmin) vs. DSS. Find Vmin by lowering dispense time at a fixed DR of 0.8 cc. Repeat this for the DSS value found in Step 2, plus 250 rpm increments up to 3000 rpm for 15-20 cps and 4000 rpm for 30-60 cps. In most cases, you will find that Vmin will go down considerably as you increase DSS. For SPR508 (13.8 cps), Vmin = 2.2 cc @1200, and Vmin = 1.36 cc @2400 rpm. Plot Vmin vs. DSS. Pick a dispense volume and a DSS from the graph. Pick a dispense value at least 0.2 cc larger than Vmin as a safety margin.

Fig. 10. Use the six-step checklist in this article to generate your own uniformity chart similar to that shown here.

Fig. 11. Use this sample chart to optimize your own spin coating process.

Step 4: Process optimization should be established using a 5 factorial DOE for improved uniformity. Optimize Tr, Tw, DR, DSS and EFR in order to get the most uniform film profile. In our study of SPR508 (13.9 cps) and JSR300 (55 cps) on the SVG 90SE track, the same DR (0.8 cc/sec) and exhaust flow rate (200 liters/min) values were found to be optimum for both resists. Tr and Tw needed to be optimized carefully, whereas the dependence on DSS was very weak. In many cases, a simple DOE involving only Tr and Tw may suffice to optimize the process. The higher the viscosity, the more variables that may need to be included in the DOE. Graph the results as illustrated in Figures 10 and 11 to see the optimum setpoints and process latitudes.

Step 5: Process stability -- Coat 100 wafers using the optimized recipe determined in Step 4. Measure film thickness profiles. Analyze the data statistically, and plot mean thickness and uniformity as a function of wafer number. Establish control charts to investigate the long-term stability.

Step 6: Defect investigation -- Investigate the coated wafers for macroscopic visual defects under a monochromatic light. Use a particle measurement instrument to quantify the detection of microscopic defects.

Summary

By following the proper optimization of process parameters, resist usage has been reduced on production wafers to the 1.5-2.5 cc range for i-line (mid-viscosity) resists and 1.0-1.5 cc range for DUV (low-viscosity) resists. This program has resulted in a >= 45% reduction in resist consumption, allowing a cost reduction of > =$1 million/yr for semiconductor fabs. An additional benefit of this reduction program has been the ability to extend operational times between preventive maintenance, since less residual material is coated onto the catch-cup of the coater module.

Further reduction will require a radical new approach to controlling mass transfer and evaporation mechanisms that ultimately control the spreading and drying of resist during the coating process. Variable rate pumps are being evaluated that ensure that the initial film created on the surface of the wafer is uniform. In addition, a solvent mediated atmosphere created over the wafer surface during coating would serve to improve wet-ability at the liquid-solid interface and eliminate surface tension variations across the wafer's surface.² This type of approach would render the coating process insensitive to environmental variations (including pressure, temperature and %RH) and has already demonstrated good results with <1.0 cc of dispensed resist.

References

1. D.E. Bornside, C.W. Macosko and L.E. Scriven, J. Appl. Phys., 66(11), 5185 (1989) 9.

2. U.S. Patent #5,670,210.