Abhay Ramrao Deshmukh, National Semiconductor

The oft-repeated mantra about reducing costs took on new meaning during the prolonged downturn that afflicted the semiconductor industry. Accordingly, cost-reduction projects have risen to the top of many manufacturers' must-do lists or have been adopted by companies for the first time. The drive to reduce costs has been emphasized through meetings, e-mails, special announcements, and even layoffs. Projects that can save money and add dollars to the bottom line are finding ever-greater favor among corporate executives. In virtually every company, the hunt to cut costs is on.

Process engineers are faced with the challenge of making processes even more efficient, effective, and cheaper. However, project qualification and implementation, with cost cutting as a primary objective, involve the careful screening of many factors, including device layout, functionality, and applications. Fabs manufacture a wide range of products, and the changes made to one complicated device or processing technology cannot necessarily be transferred to others. If process engineers are unaware that a device or process is sensitive to a certain change they have made, the change may come back to haunt them. Hence, understanding cost cutting and its impact on complex process changes is crucial. The optimal solution involves balancing changes against presumed cost-reduction benefits.

This article describes a designed experiment that was performed at National Semiconductor (Arlington, TX) to lower costs associated with the spin-on-glass (SOG) deposition process. It discusses the role of SOG in semiconductor manufacturing and presents a case study that enabled the fab to reduce SOG consumption and costs.

The Role of SOG in IC Manufacturing

SOG, which is used primarily to planarize interlevel dielectric to avoid poor step coverage of deposited metal or voids in deposited dielectric films, is one of the most expensive materials employed in semiconductor manufacturing. In addition, SOG solution has a strong tendency to form particles, while excess material spun onto the walls of process tools dries quickly and results in particulation. Eliminating SOG-based particles from the insides of process tools involves frequent chamber cleans, line changes, and exhaust cleans—not to mention recurring tool qualifications. All of these requirements make SOG deposition a prime example of a costly and inefficient process. Thus, efforts to improve chemical utilization and equipment performance can lead to substantial cost savings.

The amount of planarization required on the wafer surface determines the optimum thickness. SOG thickness, in turn, is contingent on metal-line widths and spaces and the aspect ratios of next-level vias. Planarization depends on SOG solution properties, application methods, and underlying topography. The thicker the SOG, the better the planarization. However, thicker films have a greater tendency to crack when cured.

Typically, SOG is applied in two or more separate thin coats. Each coat is followed by a baking step and sometimes a curing step, after which the next coat is applied. Multilayer films deposited using this iterative approach have much better cracking resistance than a film of the same thickness produced in a single coat-bake-cure cycle. Multicoat films are typically built up to thicknesses of 3000–7000 Å.

Figure 1: SOG film's raised-center pattern, which results from it being applied as a spin-on process.

Since SOG films play a critical role in device planarization, it is of utmost importance to achieve the highest possible degree of within-wafer film uniformity (<0.5%). However, because it is applied in a spin-on process, the films naturally exhibit a raised-center pattern, as shown in the illustration in Figure 1. Given SOG's as-deposited hat-shaped film distribution, the planarization process is highly sensitive to film nonuniformities resulting from such process inconsistencies as etch-rate variations, SOG/CVD-oxide selectivity, and changing etch-chamber conditions.

SOG Deposition Process

As the name implies, SOG is a spin-on process in which fluid is deposited onto a wafer or substrate. The deposition process is divided into four stages: deposition, spin-up, thinning, and evaporation. The first stage involves depositing fluid onto the wafer using a dispense nozzle. In most cases, much more coating solution is deposited onto the substrate surface than is ultimately required to achieve proper film thickness. In stage two, the spin-up step, the substrate is spun until it reaches its final rotation speed. During this process, fluid is aggressively expelled from the wafer surface by the rotational motion. Eventually, the fluid is thin enough to rotate completely with the wafer. In the third and fourth stages, the thickness of the film coat is decreased through thinning and the evaporation of solvents.

The dispense volume, pressure, and rate at which SOG films are deposited influence the characteristics of these films. In particular, dispense volume is a critical parameter because it affects film thickness and, ultimately, the planarization process. In the experiment described in this article, these three parameters were studied with the objective of minimizing SOG dispense volume without impinging on final film properties.

Since the planarization requirements for different process flows and devices vary, devices with the most challenging topographies and planarization needs were selected for this study.

Experimental Setup

The amount of SOG used during the deposition step is conditioned by the dispense volume required to achieve good planarization and the efficient use of SOG chemicals—that is, the need to reduce waste. To single out these factors, this study was divided into two components: dispense-volume reduction and waste reduction.

The experiment used an 817 track system from Semix (Fremont, CA) that was equipped with a Model 550 pump and filtration system from Integrated Designs (Carrollton, TX). One coat of Accuglass 211 SOG from Honeywell Electronic Materials (Sunnyvale, CA) was deposited. Each wafer was subsequently baked at 300°, 200°, and 100°C for 30 seconds. A Prometrix analysis tool from KLA-Tencor (San Jose) was used to measure as-deposited film thickness and within-wafer film-thickness uniformity.

Figure 2: Linear relationship between pump pressure and flow rate.

The first step in the experiment was to establish the SOG dispense flow rate, since flow rate determines the dispense characteristics of the material in the first stage of the deposition process. To determine the flow rate, it was necessary to understand the characteristics of the M550 pump. Dispense volume and flow rate were controlled using Chemnet from Integrated Designs, pump software that enables users to achieve accurate recipe parameter control. The flow rate was established by increasing pump pressure from 3.5 to 6 psi in seven increments while collecting dispense volume data for 5 seconds. Figure 2 shows the linear relationship between pump pressure and flow rate.

The Effect of Reducing Dispense Volume on Thickness and Uniformity

The goal of the experiment was to determine the minimum possible dispense volume with which a highly uniform layer of SOG of desired thickness can be deposited. The pump parameters of dispense volume, pressure, and flow rate were used as input factors, while as-deposited film thickness and film-thickness uniformity were the primary response factors. Dispense pressure and volume were plotted and prediction profiles were generated to explore film thickness and film-thickness uniformity behavior.

Based on an understanding of the pump's flow versus pressure characteristics derived from Figure 2, pump pressure was changed to control the dispense flow rate. With the dispense valve kept completely open, the SOG dispense flow rate was changed from 0.8 to 1.6 ml/sec in four 0.2-ml/sec increments, while the dispense volume remained constant at 1.75 ml/dispense. The same procedure was repeated at dispense volumes of 0.75, 1.0, 1.25, 1.50, and 1.75 ml. The experimental conditions are outlined in Table I.

DispenseVolume (ml) Dispense Pressure (psi) DispenseFlow Rate (ml/sec) FilmThickness (Å) Within-WaferFilm Uniformity (Å) 0.75 3.5 0.80 1695 12.50 0.75 4.0 1.00 1693 9.50 0.75  5.0  1.25  1716  5.50 0.75  6.0  1.50  1732  13.00 1.00  3.5  0.80  1652  20.00 1.00  4.0  1.00  1669  10.25 1.00  5.0  1.25  1666  11.00 1.00  6.0  1.50  1701  6.00 1.25  3.5  0.80  1633  17.00 1.25  4.0  1.00  1642  6.25 1.25  5.0  1.25  1637  17.00 1.25  6.0  1.50  1663  9.00 1.50  3.5  0.80  1608  11.75 1.50  4.0  1.00  1622  8.50 1.50  5.0  1.25  1628  10.90 1.50  6.0  1.50  1641  11.00 1.75  3.5  0.80  1602  12.50 1.75  4.0  1.00  1610  12.00 1.75  5.0  1.25  1604  11.50 1.75  6.0  1.50  1624  11.00

Table I: Design-of-experiment data with dispense volume, pressure, and flow rate as input variables and film thickness and uniformity as response factors.

To perform response surface analysis, prediction profiles, illustrated in Figures 3 and 4, were generated using JMP software from SAS Institute (Cary, NC). Figure 3 shows the prediction profile of SOG thickness and film uniformity as a function of dispense volume and pressure. Film thickness was found to be inversely proportional to dispense volume and directly proportional to dispense pressure. Table I indicates that as dispense volume dropped from 1.75 to 0.75 ml at a dispense pressure of 6 psi, film thickness increased by 108 Å. And as dispense pressure increased from 3.5 to 6 psi at all dispense volumes, thickness increased. Consequently, it was determined that the quantity of SOG used in the deposition process could be reduced without jeopardizing film thickness.

Figure 3: Prediction profile of SOG thickness and film uniformity as a function of dispense volume and pressure. Film thickness was found to be inversely proportional to dispense volume and directly proportional to dispense pressure. Within-wafer film-thickness uniformity is not greatly affected by either dispense volume or pressure.

The experiment also demonstrated that dispense pressure and volume had very little effect on film-thickness uniformity. While uniformity improved slightly as dispense pressure increased from 3.5 to 4.83 psi, it was not greatly affected through the range of input parameters covered in this experiment.

Utilizing SOG Effectively to Reduce Waste

After the SOG deposition process had been analyzed and reviewed, two additional areas were investigated to lower consumption and costs: SOG predispense and bottle utilization.

Figure 4: Combined effect of dispense volume and pressure on SOG thickness.

SOG Predispense. Since the evaporation of solvents such as isopropyl alcohol (IPA) causes particles to form in SOG, a predispense step is performed when processing begins. The primary purpose of this step is to clean the dispense nozzle and SOG line in order to flush out SOG residues. Dispense lines are cleaned by flushing out material at the beginning of each dispense step. Predispense amounts vary according to process sensitivities, tool idle times, and the general cleanliness of lines, nozzles, and the exhaust system. Normally, between 0.5 and 1.0 ml of SOG material is used to perform the predispense step.

In this study, the effect of the predispense amount of SOG on particles added was studied as a function of tool idle time. After every four hours of idle time, varying amounts of SOG were used during predispense to flush out the components of the system. Then all parts of the dispense station were cleaned with IPA except for the dispense nozzle. A test wafer was then scanned using a 6200 analysis tool from KLA-Tencor to collect preprocessing particle data, following which it was processed in the tool. After processing, the wafer was scanned again to determine postprocessing particle performance. Particle adders were calculated by subtracting the number of preprocessing particles from the number of postprocessing particles. As presented in Table II, the largest number of particle adders (i.e., 37) was observed when the predispense step was not performed. However, when the predispense amount of SOG was as little as 0.25 ml, particle adders decreased to 17—a substantial improvement in particle performance. This test indicated that while predispense is important, the amount of fluid used can be reduced substantially to avoid SOG waste.

Test Predispense Amount (ml) Particle Adders 1 0 37 2 0.25 17 3 0.50 10 4 0.75 15 5 1.00 14

Table II: Effect of SOG predispense amount on particulate performance.

Utilization of SOG Bottles. Since SOG tends to settle at the bottom of bottles, creating particle issues, most dispense systems mandate that material at the bottom of the bottle be left unused. That unused portion can amount to as much as 30% of the bottle's overall contents, resulting in frequent bottle changes and tool qualifications.

The pump and filtration unit used in this study dispenses very accurate and repeatable amounts of chemicals using pressure-on-demand technology. Detailed software drives the sophisticated electronics that control the pump, which uses helium pressure to push chemicals through the system and onto the wafer. The pump is designed to keep process chemicals under a helium blanket to prevent crystallization, which is indispensable in SOG coating processes. The unit also has a 0.04-µm in-line filter, which helps to keep the line clean and free of particles. These features eliminate the need to underutilize SOG, improving utilization by 15–30%.

Conclusion

SOG deposition was an excellent candidate for a cost-saving study because it is expensive and has stringent processing requirements. The study demonstrated that dispense volume can be reduced by optimizing dispense flow rate and pressure. It was shown that in the manufacture of devices with challenging topographies and planarization needs, the cost of consumption can be reduced by optimizing the dispense volume. The study also showed that film-thickness uniformity is independent of dispense volume or pressure.

Further tests proved that the use of SOG during predispense to flush out system components can be reduced to a minimum and that the consumption of additional amounts of the material has little benefit. The pump and filtration system used in this study enables users to consume the full contents of SOG bottles.

Thus, by minimizing the SOG dispense volume, optimizing the predispense step, and minimizing material waste, very significant cost savings of more than 50% can be attained.

Abhay Ramrao Deshmukh is a process engineer at National Semiconductor in Arlington, TX. He has worked at the company for four years in the areas of PVD systems and SOG planarization. He received an MS in materials sciences from the South Dakota School of Mines and Technology in Rapid City. (Deshmukh can be reached at 817/557-7602 or abhay.r.deshmukh@nsc.com.)