Volume consumer markets are changing MEMS manufacturing
ERIC MOUNIER and J??R??ME BARON, Yole D??veloppement, Lyon, France
Changes in the way MEMS are manufactured are primarily being driven by high volume consumer MEMS markets.
High volume consumer MEMS markets are driving manufacturing changes across the MEMS industry, pushing companies to drive down die size, closely connect multiple die in combinations, and move towards an efficient infrastructure for system-level integration of high value solutions for final OEM customers. That means new demands ???and new opportunities???for tighter process controls, more integrated wafer-level packaging technologies, more standard process and package platforms, and more software for easy-to-use integrated sensor solutions.
Figure 1. The MEMS market is increasingly dominated by consumer applications, which inescapably means high volumes, short product cycles times, and pressure for low costs. Source: Yole Developpement.
While the MEMS market still contains a highly diverse range of products, with plenty of specialty niches and high performance applications, it's also increasingly a volume consumer business. Consumer applications accounted for more than 50% of total MEMS industry revenue in 2011 (Fig. 1). And the four main devices for consumer mobile applications-- accelerometers, gyroscopes, magnetometers and microphones???accounted for more than 50% of all MEMS units shipped last year (including automotive). Volumes will continue to expand, as smart phone shipments are likely to keep growing strongly, from 300 million units in 2011 to more than 800 million by 2016, and MEMS content per phone is also growing rapidly. More of those MEMS devices will be combined into multisensor modules, both to reduce costs and to improve performance, as more accurate location and navigation information becomes a key distinguisher for smart phones. We expect these combo units will steadily replace discrete sensors and account for most of the growth in the inertial sensor market going forward, to become a $1.7 billion opportunity by 2017.
Packaging innovations and production volumes drive die shrinks
Key to driving down costs to enable the growth of the consumer MEMS market is the rapid scaling down of die and package size. Over the last three to four years, MEMS 3-axis accelerometer die have shrunk from 12mm2 down to 2mm2. The next generation is headed to 1mm2 die size. That has helped push production cost of the packaged and tested consumer accelerometers sharply down from nearly $.40 in 2008 to $.25 by 2011. Gyroscopes are similarly shrinking fast. Typical 2-axis gyroscope die were 6mm2-8mm2 in 2007. Now InvenSense's 3-axis gyro is ~4mm2. Manufacturing cost has come down accordingly, with the current 3-axis packaged gyros costing the same $0.75-$0.85 to produce as did old 2-axis versions three years ago.
Production volumes are one key contributor to driving down die size and costs. Though wafer volumes with the tiny MEMS die of course remain small compared to CMOS, MEMS wafer volumes are getting big enough to drive tighter manufacturing controls that allow driving down the feature size. Robert Bosch credits its consumer pressure sensor production for accelerating its tight control of its manufacturing process, as getting to multimillion unit volumes in months instead of years allows it to push the limits of size and thickness much faster, which it has then transferred to its automotive products. STMicroelectronics famously credits the high volumes of its single process platform used for all its accelerometers and gyroscopes for driving the tight process control to reduce structure size and maintain consistent performance across units.
Now new packaging approaches are starting to play an increasing role in shrinking the devices. The increasing use of metal-to-metal sealing to replace glass frit bonding frames will significantly reduce the die space required. SystemPlus' reverse engineering of STMicroelectronics latest 3-axis accelerometer, for example, shows it is using gold eutectic sealing instead of glass frit to shrink packaged die size from 4.7mm2 in the last generation down to 2.1mm2. Wafer-level packaging and through-silicon or through-glass vias are poised to reduce the contact pad area to shrink the die size even more. ST has recently launched an accelerometer with TSV, for example. By eliminating the area once used for I/O pads, the TSV process allows the MEMS die area to be reduced by ~25% compared to the usual accelerometer. It adds major manufacturing changes that increase the final wafer cost by more than $100, but the reduction in die size still makes the final die cost competitive.
Figure 2. Demand for smaller, lower cost die, and faster handling of more complex multisensor data will drive MEMS packaging towards more value at the wafer level, with more TSVs and monolithic integration. Source: Yole Developpement.
Combo sensors create new demands
The coming industry transition to combination sensors, putting two or more MEMS sensors in a single module, sharing one ASIC and one package, will also make more complex demands on backend processes (Fig. 2). Packaging, assembly and test currently account for some 35%-45% of the typical manufacturing cost of a MEMS accelerometer, more than either the MEMS die or the ASIC, and the added complexity of the combo modules will likely move even more of the value to these manufacturing steps. Effectively managing increasing volumes of sensor data from the combos, such as to reduce sensor drift for more accurate position sensing, will require shorter, faster connections between the components than the current standard wire bonding, likely driving the modules towards interposer and TSV solutions. That means the modules will need to be assembled with high yields from known good die to be economic, and then all six or nine or ten sensor axes tested and cross-calibrated. Manufacturers will also need to figure out how to assure users of second sources for the complex multi- component systems. Fig. 3 shows one example.
Figure 3. Intertial sensors combo stacked in SiP module on organic laminate substrate. Courtesy of STMicroelectronics.
Cost and volumes drive MEMS
While most successful MEMS IDMS and foundries have long worked to re-use the same front-end process module for different devices as much as possible to speed time to market and time to yield, volume consumer markets are driving the sector towards standard package platforms as well, as these consumer markets do not have the luxury of time or cost to support development of fully custom packages. Naturally the wide range of different MEMS devices will maintain their different needs, but several different types of standard platforms, driven by big players in big markets, will likely develop, such as WLP and TSV interconnects, or SiP modules with micro leadframes, or chip arrays using molded or cavity packaging (Fig. 4). For example, microphone packaging from major suppliers has all now settled around the same general platform MEMS and ASIC wirebonded in a SiP module on BGA/LGA laminate PCB substrate with an air access hole, under a metal lid.
Figure 4. The wide variety of MEMS devices are now packaged in a diverse variety of package types.
Substrate suppliers and packaging subcontractors from the IC world are looking at developing libraries of building blocks or common platforms to ease development of MEMS packages for common product families. Unimicron in Taiwan, for example, is looking at giving a second life to its BGA or LGA substrate technology by developing a hole-making process to make the substrates suitable for multiple various MEMS applications. Tong Hsing Electronic Industries (THEIL) is similarly looking to develop a portfolio of re-usable blocks to speed development of MEMS packages for common product families.
Testing and calibration of the sensors ???and increasingly cross calibration of sensors with other kinds of sensors perhaps from other suppliers???are also becoming more standardized to reduce costs at higher volumes. Test already accounts for some 18% of manufacturing costs, and testing and cross-calibrating more different stimuli in combination is becoming more complicated and requiring more costly equipment, with more sites in parallel for higher throughput. Companies like STMicroelectronics, which traditionally developed their own calibration equipment in house, are increasingly turning to buying commercial equipment from vendors like SPEA. Higher volumes and cost pressures will continue to push the industry towards the use of more standard high throughput commercial equipment, as well as drive efforts towards better design for test, more wafer level testing, and more testing service providers.
Higher volumes, more common platforms
This maturing of the MEMS business to a higher volume, less custom business may open opportunities for players from the IC world, who have so far played little part in this specialty, artisanal business. With MEMS volumes getting to levels where statistical process control starts to make sense, GLOBALFOUNDRIES claims that using CMOS-style monitoring, feedback and control at every step enabled it to go from equipment install to qualified products for volume production in what is, for the MEMS industry, a lightening pace of under two years. Fabless InvenSense has ramped its volume gyroscope business at TSMC and GLOBALFOUNDRIES. In a year when the overall MEMS foundry business is significantly lagging the 17% expansion of the industry overall, with only ~5% growth, players from the CMOS side did relatively well. TSMC saw 15% growth, and Xfab 33%.
ERIC MOUNIER is senior analyst, MEMS devices and technologies, Yole D??veloppement, Mounier@yole.fr, and J??R??ME BARON, business unit manager, advanced packaging, Yole D??veloppement, firstname.lastname@example.org.
Solid State Technology, Volume 55, Issue 6, July 2012