Batch-deposition Module Cuts Downtime And Material Waste
With the average selling price of many compound semiconductor devices sliding by up to 20% year after year, it is becoming ever more critical for chip makers to improve margins by maximizing wafer yield, throughput and manufacturing uptime, while reducing the amount of scrap material produced.
Investment in a new, larger tool to increase manufacturing capacity is part of the solution, but the economics of nearly all device production requires attention to a myriad of variables in cost-of-ownership calculations. As might be expected, process system performance provides many of the critical variables. Obtaining a favorable cost analysis for tool operation implies systems that guarantee long uptimes and high throughput while maintaining a low level of scrap. Critical though this data might be for making decisions relating to production and capacity, collecting it is usually a tedious process and as a result such information is often unavailable.
Several years ago, Unaxis introduced a production batch plasma-enhanced chemical vapor deposition (PECVD) module for silicon nitride (SiNx) deposition. Using the same basic technology and process regimes of earlier systems, a new, larger module has now proved itself under high-volume production conditions, and the data presented here confirm its usefulness.
Controlling film thickness
PECVD SiNx is used in many different III-V-based devices such as MESFETs, HBTs and HEMTs in the roles of passivation and encapsulation (figure 1). In particular, the large dielectric constant of SiNx makes it attractive for use as an intermetallic insulator material in metal-insulator-metal (MIM) capacitors. To meet circuit and device-design criteria as a capacitor dielectric, SiNx film thickness and insulator properties need to be tightly controlled.
A SiNx deposition process was developed for GaAs monolithic microwave integrated circuits (MMICs) and used in high-throughput batch PECVD production at RF Micro Devices. PECVD is well suited to the low-temperature constraints of GaAs-device manufacturing, as the technique allows growth of high-quality SiNx films below 400 °C with deposition rates of approximately 100 Å/min.
The PECVD system uses a conventional parallel-plate configuration operating at 13.56 MHz. Dilute silane in nitrogen and ammonia are combined in a plasma to form the SiNx film. Nitrogen is used as a carrier gas, and it is also introduced as a separate gas in the deposition process. It is possible to control the film stress over a stress range of about -300 (compressive) to 300 MPa (tensile) by modifying the plasma chemistry .
Controlling film stress is critical because it strongly influences the mechanical integrity and electrical characteristics of the manufactured device. It is is also useful for maintaining either low composite stress in device layers or for introducing a desired level of stress to define specific film properties. A stress of around zero ensures that even relatively thick SiNx films will have a small absolute stress.
Film specifications for the batch PECVD production solution exceed the requirements for most MMIC applications. In general, SiNx film specifications cover within-wafer, wafer-to-wafer and batch-to-batch film thicknesses, since a highly uniform target film thickness is essential for meeting capacitance design criteria. Statistically speaking, the standard deviation of a film thickness must be within 2.5% of the specified target.
To establish whether these targets had been successfully met, more than 1700 samples of film thickness were collected over a more than a year using a spectroscopic reflectometer (figure 2). The data shown correspond to SiNx films deposited on five different systems and have been normalized to the target thicknesses. The narrow, near-Gaussian distribution levels reveal a normalized standard deviation of 0.0097, and this confirms a tight distribution around the target film thickness.
Similar results to this have been found with three other film-thickness goals that spanned a range of between one and 10 times the thickness of the thinnest film. For proprietary reasons, the film thicknesses shown in figure 3 have been normalized to the thinnest film produced (that is, film thicknesses were two, four, and nine times that of the thinnest film).
Wafer-to-wafer thickness reproducibility was also demonstrated to be within the 2.5% target window (figure 3). More important was that the within-wafer uniformity over the measured production runs also met the required level of reproducibility. All of this was achieved, in part, because the reactor encourages deposition uniformity owing to the careful design of a showerhead for process-gas introduction and the use of a unique pumping manifold.
Reproducible film quality
Being able to reproduce high-quality film is as important as meeting film-thickness targets. In the case study presented in figure 4, the films were used as capacitor dielectrics. Wafer-to-wafer capacitance-density reproducibility is presented for three different films and, as with the film-thickness measurements, the capacitance-density figures are normalized to the smallest value. Again, the reproducibility surpasses the 2.5% specification goal for both capacitance density and for wafer-to-wafer uniformity.
By achieving the desired process stability, a high yield is ensured. After device performance, yield improvement is often considered to be the next most critical goal in a manufacturing environment. Device-function requirements, along with process capability determine the level of qualified product. Reproducibility within upper and lower bounds, as described by statistical process control, is determined by many variables, including maintenance schedules and process stability (of gas flow, pressure, RF power and temperature, for example).
In this case, the reproducibility is attributed to a highly robust system and routine maintenance procedures. The tighter the capacitance distribution, the better the final product yield will be of devices with crucial characteristics such as high efficiency and high gain.
The batch deposition module used in this work was intended not only to provide SiNx films of the required quality, but also to increase throughput. To do this, it uses an appreciably larger chamber than is typical, and can, therefore, accommodate significantly more wafers in each batch. The two standard batch sizes are eight 100 mm-diameter wafers and five 150 mm-diameter wafers. Available on a cluster tool platform, the large-area PECVD batch system offers fully automated, cassette-to-cassette processing, and batches are loaded from the cassettes using a rotating indexer in the process chamber.
Equally important to the throughput figures used in cost-of-ownership calculations is the system s operational uptime. Figure 5 shows the uptime of four different batch PECVD systems recorded over a period of a year. Except for a single month on system 4, where the operational uptime dipped to 93%, the uptime performances were consistently above 95% with an average uptime of more than 98%. Scheduled maintenance time is counted as downtime, as described in the international semiconductor guidelines provided by Semi E10-0701. This preventative maintenance averaged approximately 2% of measured downtime per system over the reviewed period.
Interpreting these uptime figures requires at least a rudimentary understanding of the maintenance cycle and practices involved when using a batch deposition module. The PECVD reactor includes several features to maintain system cleanliness that all serve to enhance yield and minimize system downtime. For example, the chamber walls and the upper gas-distribution electrode of the reactor are heated to minimize particulate formation during SiNx deposition. As might be expected, SiNx is deposited on the wafer and also on exposed surfaces in the chamber. If this material were allowed to build up indefinitely, issues with process repeatability and particulate control would arise.
To help maintain consistent levels of efficiency, the standard PECVD SiNx process incorporates both a regular in situ clean and a manual cleaning process. The in situ clean is performed more frequently (after approximately 1.5 μm of deposition) and is based on a fluorine plasma etch.
To achieve and consistently maintain the reactor in a clean, known state, an automatic plasma etch-back sequence is integrated with an optical emission spectrometer (OES). The OES ensures that effective in situ cleaning is completed in the shortest possible time. Typical in situ cleaning proceeds at a rate of approximately 2500 Å/min and thus cleaning takes around six minutes or about 4% of the deposition time. As a result, manual cleaning has to be carried out only once a year.
A number of important features are available on the batch PECVD system to improve system maintainability and uptime. These were all implemented based on experience with the standard PECVD system and include clean-chamber design. Modifications were made to minimize the number of parts of the reactor that tend to collect deposited material and become difficult to clean. A heated top electrode improves processing ability and system cleanliness, while port access to the main pump manifold allows for easier manual cleaning of the main vacuum manifold.
Now proved at various customer production sites, we conclude that the batch PECVD system provides advanced III-V device manufacturers with a high-throughput solution for depositing high-quality SiNx at low process temperatures. Extensive testing has demonstrated that these results are not only repeatable for production with 150 mm-diameter substrates, but they are also achieved with uptimes of approximately 98% and with negligible scrap generation.
 K D Mackenzie, B Reelfs, M Devre, R Westerman and D J Johnson 2004 Proc. CS Mantech Technical Digest pp319-322.