by Tony McKie, memsstar
Dry-etch process technology enables yield growth on advanced devices to create a fundamentally low cost base for MEMS development and production. Failure to address yield and production flexibility issues while expanding throughput to satisfy increasing MEMS demand will not allow MEMS industry participants to realize the full market potential of increasingly sophisticated MEMS devices.
Introduction: Opportunities and challenges
The MEMS industry has identified tremendous opportunities in applications such as mobile phones, game consoles, automobiles, and many other consumer products. Already, MEMS devices such as tiny surface-mount accelerometers and gyroscopes are enabling innovative features such as automatic screen-orientation detection for camera-phone handsets and motion-sensitive game controllers offering unprecedented interactivity. Other advances made possible by MEMS include ultra-miniature silicon microphones combining tiny dimensions with high dynamic performance for use in cell phones, headsets, and similar space-constrained applications.
To fulfill these major high-volume opportunities for MEMS manufacturers and realize the expected growth potential, device producers must be able to manufacture in high volume, achieve competitive prices compatible with consumer applications, and deliver more advanced and high-performing MEMS products in the future.
This combination of objectives presents a significant challenge. An intuitive response is to increase production -- for example, by using batch-processing techniques -- to deliver high volumes while at the same time realizing economies of scale. However, batch processing does not provide the ability to produce more advanced structures. In practice, it will also not deliver the anticipated improvements in throughput and cost, when all of its associated limitations are taken into account. Instead, MEMS manufacturers must seek improvements on a more fundamental level.
Some valuable pointers can be gained by considering the way semiconductor wafer-processing practices have advanced, since MEMS wafer processing has much in common with CMOS fabrication. Early semiconductor processes such as etching relied on wet chemistry, similar to first-generation MEMS release processing. However, wet chemistry has given way to dry processing, in the semiconductor industry, as shrinking process geometries have enabled progressively smaller circuit dimensions.
This transition is now evident in the MEMS industry as device manufacturers adopt a process such as vapor-phase release etching to achieve process enhancements that will enable smaller and more complex MEMS structures. Already, a number of process and equipment developers are offering dry etch to their customers, and several such advanced processes are now active worldwide.
|Figure 1. Poly-bridge with aluminum contacts/HF etch.|
Evolving MEMS processes
Release etching is the final stage of fabrication for surface micro-machined MEMS devices. This process releases the structure from the surrounding sacrificial material. As a replacement for release etching with acid, vapor-phase release processing, using anhydrous HF or XeF2 as the release agent, improves process control and increases selectivity. In addition, since the anhydrous release agent does not attack materials such as metals (Figure 1), vapor-phase release also provides greater freedom for device designers to build structures using materials that would not be compatible with a wet-chemical process.
The enhanced flexibility and control of vapor-phase release etching also allows fine tuning to deliver the best results for a given combination of process, device architecture, and structural and sacrificial materials, such as shown in Figure 2. For example, successful oxide release using anhydrous HF is achieved through accurate control of H2O vapor, as the catalyst for the reaction between the anhydrous HF and the oxide material, to maintain the desired etch rate. The memsstar vapor-phase release-etching processes permit the etch rate to be reduced or increased to achieve the best possible throughput, without compromising factors affecting yield, such as selectivity.
|Figure 2. RF-MEMS variable capacitor. Courtesy of NXP Semiconductor.|
Following the semiconductor industry's lead in moving to vapor-phase processing delivers several advantages for MEMS manufacturers. These include helping to increase yields and improve device performance and reliability and by acting as an enabler for innovative new structures.
As new markets and applications for MEMS emerge, and demand continues to grow for higher volumes and new types of products, device manufacturers can draw many more lessons from the semiconductor industry. Batch processing, for example, has recently been proposed to help speed up MEMS manufacturing and reduce unit costs. Some MEMS manufacturing equipment now on the market is capable of processing up to 25 wafers simultaneously.
However, although semiconductor manufacturing has successfully adopted vapor-phase processes, batch techniques have been tried and -- in most cases -- discarded. The majority of today's semiconductor wafer-processing lines are set up to process only one wafer in a chamber at a time. Advantages include increased repeatability and faster cycle times for each process. MEMS process engineers can learn from this trend.
When a batch of several wafers is loaded into a process such as vapor-phase etching, the time to complete the process is actually significantly longer compared to single-wafer etching. Release etching for a batch of 25 MEMS wafers, for example, can require a total cycle time measured in hours. In contrast, the process can be completed within a few minutes when applied to a single wafer.
In addition, batch processing is known to result in significant wafer-to-wafer variability, leading to relatively large numbers of rejected components. As MEMS device structures evolve toward finer dimensions, the tolerances associated with key parameters such as dimension control and film thickness are becoming significantly finer. Indeed, some of the latest MEMS designs now entering production cannot be produced satisfactorily using batch processing because wafer-to-wafer variability falls outside acceptable statistical limits. This is manifested as unpredictable yield, which prevents cost-effective manufacture.
In practice there are also loading effects associated with batch processing. This means that process performance will vary depending on whether all the wafer locations are filled, or whether a smaller number of wafers are presented for processing. If only 10 wafers are loaded into a process characterized for 25 units, the properties of the wafers emerging will differ from the expected results. Appreciable loading effects more or less force manufacturers to process only complete batches, which can slow down the production flow. Even then, the results can display noticeable variation, as already discussed. It is also extremely difficult to perform custom tasks on individual wafers, if required, in the context of a batch process.
The high wafer-to-wafer variability experienced with a batch process also effectively increases overheads such as materials consumption and the costs associated with testing. Rather than enabling manufacturers to benefit from economies of scale, therefore, batch processes having relatively poor process control incur extra costs to build and test large numbers of devices that will ultimately be scrapped.
Another factor to take into account when evaluating batch processing is the high cumulative value of the work in progress at any one time. This not only represents a significant investment in inventory, passing through the factory at a slow speed, but also exposes a relatively large number of wafers -- each containing many individual components -- to risk of damage. If a process is set up incorrectly, or is impaired by an event such as power outage or temporary loss of process control, a large number of wafers may become scrap simultaneously. Also, if a number of wafers are held up (e.g., waiting for other wafers to arrive to make up a full batch), these are at risk of damage or contamination that also may require one or more wafers to be scrapped.
Historically, batch processes have proved difficult and time consuming to qualify. There are many locations within the 3D space inside a multi-wafer processing chamber that must be characterized before the process can claim to be fully developed. Inaccuracy will compromise process performance, and hence will impair the yield from every batch. However, process developers cannot afford to spend excessive time characterizing their processes, and device manufacturers cannot afford to wait for a protracted period before a satisfactory process is ready to go live. In general, developing a satisfactory batch process demands considerable time and effort. Because a single wafer can be processed to a high standard, it does not necessarily follow that this can be scaled, reliably, to enable multiple wafers to be processed on a batch basis.
Single-wafer processing overcomes the limitations of batch processing, and meets the current demands of MEMS device producers. Additionally, single-wafer processing not only enhances process control but also allows businesses to respond more quickly to market opportunities (Figure 3).
|Figure 3. An example of a successfully released MEMS structure.|
Relatively small numbers of devices can be produced cost effectively, and on a fast-turnaround basis, for research and development purposes, or for small production runs. At the current stage of development of MEMS technology, time to market is critical. Indeed, given the rapid pace of change, the high costs and long delays associated with setting up a batch process may exceed the total market opportunity of a given device.
Because single-wafer processing has been almost universally adopted within the semiconductor industry, complex and essential equipment such as wafer-handling stations are proven, reliable, and readily available. In contrast, batch handling equipment for MEMS is today highly bespoke. It is therefore relatively expensive and likely to remain so given the effectiveness of current single-wafer processes. In addition, the bespoke nature of the batch equipment effectively requires the MEMS manufacturer to enter into an equipment-development partnership with the vendor; a euphemism for accepting poor process performance for the foreseeable future.
In addition to the relatively high cost of the equipment, size can also be a disadvantage since batch equipment is inherently larger than an equivalent setup for single-wafer processing. This can be an important consideration for facilities in locations where real-estate prices are high.
The way forward
In the immediate term, as dry-etch technology continues to mature, process development for MEMS fabrication must focus on achieving further improvement in cost. Increasing yield is the best way to establish a fundamentally low cost base for MEMS fabrication, which is necessary for economical production at low unit cost in very high volumes. If this crucial groundwork is not done, batch processing will merely increase production of bad units, and therefore deliver only limited cost savings.
Recent market analysis from experts such as Yole Développement, suggests relatively flat demand for MEMS capacity in the coming months. This presents an opportunity for MEMS producers to focus on driving up yields during this period, and hence create the environment for low-cost, high-volume production in the longer term.
The performance of the semiconductor industry shows how maximizing yield holds the key to realizing full market potential. As semiconductor device designers pursued Moore's Law to increase performance and integration, improvements in process capabilities have enabled a higher proportion of good die per wafer. This has made a valuable contribution to reducing the cost of each unit produced.
MEMS producers must make the most of opportunities to learn from the semiconductor industry's prior experience. One of the most important lessons for today's process innovators is to understand why, after alternatives have been tried and discarded, single-wafer processing remains the technique of choice for high-throughput, high-yield fabrication.
The company name memsstar is a registered trademark.
Tony McKie received his Honours degree in physics from Paisley Technology College in 1987 and is general manager and co-founder at memsstar, Starlaw Park, Starlaw Road, Livingston, West Lothian, Scotland EH54 8SF; ph.: +44 1506 409163; e-mail firstname.lastname@example.org.