Wafer Bonding Enables Future RF Filter Manufacturing
The continually growing demand for faster mobile data access and smart integration strategies is creating new and greater requirements affecting RF filter designers and manufacturers. EV Group examines ways to accelerate the production of SAW-based devices through the use of new materials and packaging methods supported by advanced wafer bonding techniques. By Dr. Thomas Uhrmann, EV Group
Mobile technology has emerged as a primary engine of economic growth and transformed our everyday lives in a profound way. With each passing year, mobile technology's spread throughout the world increases - through new types of electronic devices as well as new applications. Volume shipments of smartphones are expected to reach nearly 1.8 billion annually by 2021 . Not surprisingly, global mobile traffic growth is also rising rapidly, with some estimates of traffic usage at 49 exabytes per month by 2021 . These trends are leading to growing bandwidth demands and more crowded spectrum.
The migration from 3G to 4G and 4G LTE broadband wireless technologies has enabled multimegabit bandwidth, more efficient use of the radio network, latency reduction, and improved mobility - ultimately enabling faster download speeds. This in turn is driving a dramatic increase in the need for advanced filtering technologies, with some high-end, feature-rich phones today incorporating over 50 radio frequency (RF) filters . The transition to 5G - driven not only by consumer demand for more graphic-processing-intensive applications such as augmented/virtual reality (AR/VR) but also the Internet of Things (IoT), the Tactile Internet, Industrial 2.0/IIoT, smart grid/energy and autonomous vehicles - will further drive new filter requirements . These include different frequencies (and more of them), as well as steeper skirts in individual filter bands to reduce cross-talk between the bands and improve frequency accuracy.
Figure 1.Lithium tantalate (LTA) bonded on silicon using LowTemp plasma activation (a) scanning acoustic image (b) photography of the bonded wafer pair.
RF filters need to be simultaneously smaller, cheaper and have increased functionality in order to support these growing requirements for consumer mobile devices. However, surface acoustic wave (SAW) filters are difficult to scale dimensionally due to the physical properties of the substrate material used to fabricate them. Opportunities at both the materials/substrate level as well as in packaging are emerging that can enable RF filter manufacturers to drive down RF filter costs and footprint as well as increase filter functionality. These are:
The adoption of substrates with improved electrical properties such as lithium tantalate (LiTaO3, also referred to as LTA) and lithium niobate (LiNbO3, also referred to as LN) on silicon
The adoption of wafer-level packaging to drive down costs, reduce footprint and increase device performance for improved robustness/protection from the elements or even for hermetic sealing
Wafer bonding plays an important role in enabling the integration of new materials like LTA and LN on silicon in SAW filter manufacturing. This article will explore several wafer bonding technologies that are needed for both substrate processing and packaging of LTA- and LN-on-silicon based SAW filters.
Wafer bonding considerations for new substrate combinations
Bulk LTA and LN substrates possess unique optical, piezoelectric and pyroelectric properties that make them valuable for SAW applications such as RF filters. However, LTA and LN are very expensive as well as brittle materials, which make them prone to breakage and yield loss. In addition, LTA and LN are anisotropic materials, which have different linear expansion coefficients in different directions. RF filters built with these materials have a temperature yield drift, which makes it very challenging for the filter to stay on the designated band. As a consequence, the filter chip has to be physically broad - with relatively wide spacing of the interdigitated finger structures deposited on the filter - in order to compensate for the temperature-related shift and remain on the designated band while maintaining good filtering properties with little to no signal degradation.
Figure 2. Example of a plasma-activated wafer bonding process flow
To address this thermal expansion and band drift problem, a thin layer of LTA or LN can be bonded onto a bulk silicon substrate, with the subsequent wafer stack processed, diced and packaged versus manufacturing RF filters on bulk LTA or LN substrates. Unlike LTA and LN, silicon is isotropic, whereby the substrate expands at the same rate in every direction. In a typical LTA-on-silicon stack, the LTA layer may be as thin as one micron or even less, while the silicon layer is 100 times thicker in the final filter. Representing the bigger component in the thermal expansion equation by far, the silicon stabilizes the thermal properties of the filter. This makes the filter less prone to reacting to temperature changes and parasitic effects. This allows the thickness of the filter and band selection to be made much narrower and more finely tuned, keeping the frequencies locked to a tighter band. This approach has additional cost and yield benefits. For example, since silicon is a much less expensive material compared to LTA and LN, the overall cost of the filter can be reduced. At the same time, silicon is a material that is already well understood in the wafer fab and easy to incorporate into a volume production environment.
Wafer bonding challenges
Direct wafer bonding is a bonding approach that enables the combining of two different materials with different lattices and coefficients of thermal expansion (CTEs) without any additional intermediate layers. The bonding process, which is based on chemical bonds between two surfaces that are established by elevating the temperature of the surfaces and applying pressure, can be used to enable LTA/LN on silicon. However, there are several key considerations with direct wafer bonding:
Surface roughness: excessive roughness inhibits sufficient contact of the wafers, which leads to low bond strength or no bonding at all
Cleanliness: particles on the wafer surface result in voids due to lack of surface contact in that region of the wafer
CTE Mismatch: at high bonding temperatures, CTE mismatch introduces stress that results in wafer bow and can even lead to cracks
Current methods of manufacturing LTA and LN wafers are generally less sophisticated compared to silicon wafer manufacturing. For example, bright polishing is often used instead of chemical mechanical polishing (CMP), which is insufficient for properly conditioning the wafer's surface prior to bonding. In addition, the CTE of the two materials differs significantly from silicon (by a factor of 3 in the case of LN, and by a factor of 4-6 depending upon direction in the case of LTA) . As a result, even bonding at temperatures lower than 200°C results in cracks, which cause massive yield loss. However, treating the surface of the silicon substrate with plasma prior to bonding the LN/LTA layer allows the annealing temperature to be reduced to 100°C, which in turn eliminates voids and cracking (Figure 1). In addition, a pre-cleaning step prior to plasma activation can eliminate surface roughness and particles to ensure maximum bonding yield . Plasma-activated wafer bonding thus provides an ideal process for manufacturing temperature-compensated SAW filters. Figure 2 illustrates a typical plasma-activated wafer bonding process flow.