Research on microelectromechanical system (MEMS) resonators dates back to the mid-1960s, but MEMS timing references are only now being introduced commercially. In this time, two fundamental technical challenges have been overcome: finding a stable and predictable material from which to build the resonators, and developing a sufficiently clean hermetic packaging system.
This research was motivated by a desire to replace quartz crystal resonators. Quartz crystals are manufactured in very high volume; about eight billion will be built this year. While they have many excellent properties, they also have significant limitations. For instance, quartz crystals cannot be integrated into silicon circuits, they are often large, and they have manufacturing yield and quality problems. These limitations are becoming more apparent and constraining in modern products.
The first problem MEMS researchers overcame was finding a material that was sufficiently stable to maintain a precise resonant frequency. In one year of continuous use, a typical resonator must flex a million-billion (1015) times, while only changing a few parts per million in resonant frequency. The strain levels can be large, in many cases approaching the non-linear limits of the material. In addition to quartz, the MEMS community has tested many alternatives, including metals, nitrides, oxides, silicon, polysilicon, and various combinations. Silicon has been found to be an exceptionally good material for resonators.
The toughest problem in the intervening 40 years was developing a suitable hermetic packaging technology. MEMS resonators require exceptional cleanliness because the small devices are especially sensitive to surface contamination — a single monolayer adsorbed on or desorbed from the resonator surface can push the resonant frequency out of specification. This packaging also must provide a robust mechanical cover over the MEMS structure, be small and preferably CMOS integrable, and be low cost. The recently developed encapsulation and packaging technology provides all of these and opens the way for commercial devices.
Quartz crystals generally are packaged in metal or ceramic vacuum enclosures, but MEMS resonators packaged this way usually do not show the required stability. Also, using the old packaging techniques would not leverage the MEMS strengths, namely small size, potential CMOS integrability, and low costs derived from integrated circuit manufacturing technologies. Therefore, MEMS researchers have looked beyond metal and ceramic vacuum enclosures.
Potential packaging and sealing technologies include bonded covers attached with anodic, frit glass, solder, or gold compression techniques. While covers attached this way provide mechanical protection, they do not provide clean enough environments for timing references. Bonded covers also double the thickness of the MEMS components, use significant die area for sealing rings and bondpad access, and incur significant system costs, usually greatly exceeding the cost of the resonators. For example, with bonded covers, the seal rings and bondpad routing can consume as much as 80 to 90% of the die area, and can account for over 80 to 90% of the cost of the packaged MEMS resonators.
Thin film encapsulations are an alternative to bonded covers. These usually are based on deposited layers — for example, deposited nitride or polysilicon, or on plated metal. These techniques do not burden the encapsulated part with large seal rings and restrictive bondpad layouts, but they are not intended to withstand the full pressure of plastic molding and are not generally clean enough for frequency references.
Getters have been investigated to scrub encapsulated environments and have yielded high-quality vacuums, but the added cost and complexity is significant, and integration problems exist.
The new encapsulation described here does not require sealing rings or restrictive bondpad routing, and the electrical connections are brought to the surface wherever appropriate, enabling efficient use of die area. The encapsulation is mechanically strong and survives over 100 atmospheres pressure in the plastic injection molding used in chip packaging. The encapsulation also provides a vacuum environment that is exceptionally clean and stable, and is suitable for reference oscillators. Finally, the encapsulation is highly economical.
The production process is based on common CMOS production tools and is performed in common CMOS foundries. Briefly, resonator structures are lithographically patterned and plasma-etched into silicon wafers. The structures are covered with layers of oxide and silicon, then released by removing the surrounding oxide to form freestanding resonators. The freed resonators are annealed at high temperature in an extremely clean reactive gas and sealed under an additional layer of silicon. Contacts are made to the resonators, and interconnect wiring is built with normal integrated circuit processes. The completed MEMS devices, now enclosed in small chip-level vacuum chambers, are diced from their wafers and packaged with drive electronics to form complete oscillators.
The final chip packaging deserves a special note since packaging usually dominates the cost of resonators and oscillators. The resonators with their drive circuits are molded into standard, inexpensive plastic packages developed for integrated circuits. These packages are in high-volume production, and have proven to be very reliable and economical in integrated circuit applications. This is only possible because the MEMS resonators are safely protected under durable silicon covers.
These MEMS oscillators have been tested for frequency accuracy, frequency stability, package hermeticity, temperature cycle durability, accelerated aging, and across various other performance and reliability regimes. In these and other measures, the MEMS resonators and drive circuits perform exceptionally well, meeting or exceeding the requirements for many consumer and industrial applications.
The MEMS oscillators now being sampled are the world’s smallest programmable oscillators. They meet or exceed specifications for a wide variety of applications, and they cost less to manufacture than the quartz parts they replace. They are being manufactured with high yield in standard CMOS foundries, and packaged in standard integrated circuit production lines. They will be in volume production in only a few months, and will be built into a wide range of products a few months later.