In the MMIC arena, microwave functions previously achieved with a chain of multiple single-function MMICs are being replaced with more highly integrated, application-specific, multi-function MMICs. As a result, functions recently requiring 10 to 15 individual MMICs are now being designed with as few as four MMICs.

Similarly in the system area, multiple connectorized module microwave subsystems have given way to single printed circuit board assemblies utilizing high frequency surface- mount packaged MMICs. This technology evolution significantly reduces the required skill level, dedicated resources, and cost of the microwave subsystem assembly, shifting much of the specialized technical burden to the MMIC supplier. With this shift, an optimal result requires the balance of new tradeoffs in addition to those with which system designers have become accustomed.

Performance Drivers

Customers for point-to-point systems are increasing their performance demands based on several aspects, including network upgrades through 4G. In developed areas, higher data users with spectrum intensive smart phone applications are pushing past the available capacity in the installed network base. The need to expand capacity results in required upgrades to support higher data efficiency in bits/Hz of available spectrum. This is being accomplished through a migration to packet-based architectures while using higher ordered, linear modulation schemes through 256 QAM and replacing predecessors, which used lower ordered QPSK through 64 QAM modulations. The higher ordered schemes deployed currently require more linear power (higher TX P1dB and IP3), while “green” base stations have less DC power available. In addition, while the higher ordered modulation schemes can carry more data in ideal conditions, they are not as robust in adverse weather conditions. To account for this, new systems have the added complexity of adaptive modulation techniques. The wireless link will operate at full capacity in good conditions and drop to a lower capacity but more reliable modulation, such as 16 QAM in adverse conditions. This will preserve voice transmission while slowing but preserving data access. Enabling this system level-flexibility requires a significant TX power adjustability and allows a wide dynamic range, while maintaining required performance levels.

The higher modulation schemes also require lower synthesizer phase noise, and more sensitive receivers to be able to decode the weaker signals that are closer to the thermal noise floor than ever before. In short, real demands are pushing higher TX output power and linearity at reduced DC current budgets, higher dynamic range, lower phase noise synthesizers and lower noise figure for better receiver sensitivity. This daunting set of challenges is not uncommon for system designers across multiple industries.

Business Case Drivers

The developed world’s appetite for data continues to grow, while typical smart phone plans bill for the number of minutes of talk time plus a fixed fee for access to data. This “all-you-can- eat” mentality has pushed networks to their limits for capacity as data traffic has overcome voice traffic, and some providers are rethinking the service and product offering as a result to better monetize data charges. At the same time, while the developed world is looking to increase data rates for faster and reliable access to information, the developing world is the largest area of growth in terms of added cell phone users. This high growth is typically through an enormous user base, while the average revenue producing unit/user (ARPU) is literally an order of magnitude lower than in the developed world. While this group is using less data today, higher levels of sophistication are expected quickly – such that high performance hardware is being deployed even in developing areas in order to plan for near term expansion. This combination of factors has created a heavy downward price pressure with fierce competition to enable penetration into the developing markets including China, India and others.

At the same time, there is pressure to reduce soft costs as well as direct costs by stocking fewer part types, and using fewer parts in more locations. Engineering teams across the industry have responded by migrating toward higher levels of integration and higher bandwidth devices in order to minimize the number of parts to be purchased, which has also had the effect of shrinking engineering design cycles through reuse of parts in multiple systems at multiple frequency bands.

Finally, disruptive innovation through high frequency SMT packaging development has enabled a significant paradigm shift, as companies look to outsource the build and test of millimeter-wave subsystems to low cost offshore contract manufacturers. These SMT packages have eliminated the barrier to entry created by chip and wire manufacturing technology formerly required to build millimeter-wave subsystems. This has caused an accelerating acceptance by design engineers to trade some performance due to SMT package losses for significantly lower costs.

While these SMT-based contract manufacturers may be able to build complex systems, performing RF troubleshooting of large cascaded chains of single function MMICs remains elusive in many cases. Thus, not only are small size and higher reliability enhanced by reducing parts count through higher levels of MMIC integration, but the ability to produce the parts through test and troubleshoot with high Cpk in a low cost offshore environment all critically depend on the ability to make the operation as simple as possible.

Statement of Problem

All of the above arguments culminate into the need for massive consolidation of the number of MMIC devices or packaged modules used per system via higher levels of integration with higher performance and broader bandwidth. MMICs must be available in SMT packages to enable low cost offshore CM business models to solve cost, performance and logistic problems. Again, these are a daunting but all too-commonly heard set of requirements.

Figure 1a Typical transmitter block diagram.

Design Tradeoffs and Solutions

Figures 1a and 1b show a set of block diagrams of a typical millimeter-wave transmitter and receiver for point-to-point applications, respectively. The design is the result of many tradeoffs that are completed at the system level. Tradeoffs exist between several areas.

Figure 1b Typical receiver block diagram.

Systems require a wide linear TX output power range that is achieved through strategically placing high linearity voltage variable attenuators (VVA) within the gain partitioning of the TX chain. TX output noise power and TX linearity both need to be optimized over this output power range, creating a conflict in positioning the VVAs. Moving the VVAs toward the final power amplifier will reduce TX output noise in lower power states, but system linearity is adversely impacted as the VVAs are subjected to higher power levels closer to the power amplifier, and thus create more intermodulation products.

Another significant tradeoff is that TX output power and linearity must be increased while reducing DC current. Here, the choice of MMIC design topology is critical – minimizing mixer conversion loss, relying on low noise buffer amplifiers where possible and ensuring that filtering is adequate to avoid premature linearity degradation due to undesired signals entering the TX power amplifiers. Judicious MMIC topology selection can help avoid a common mistake, boosting the RF high power amplifier requirements (and DC consumption) to increase linearity when these other topology changes can increase system linearity with a much lower DC current increase. In addition, IF gain is far more efficient than RF gain, so use of a high linearity mixer can allow lower RF gain and higher DC efficiencies as a result. Another key area is to ensure that SMT package losses are minimized to avoid the costly addition of a gain block to compensate the extra RF loss introduced. Thus another reason to reduce the total number of MMICs in an RF chain through integration is to eliminate the number of high frequency transitions and the associated loss – creating a powerful performance-based argument pushing higher integration levels.

Figure 2a Internally matched design.

Figure 2b Internally matched EM 3D analysis.

Figure 2 shows the result of a successful model development for a high frequency low loss package, that is realized as an air cavity QFN (quad flat no-lead). Note that the package was designed using EM modeling software, enabling the use of coplanar launches to achieve very low RF transition loss, under 0.4 dB at 40 GHz. A low loss air cavity QFN package is highly useful as it enables the existence of surface mountable passive components like filters, power combiners and couplers to be leveraged in even an all SMT design.

Figure 2c Resulting performance.

Filter incorporation per the block diagrams in Figure 1 help to ensure spectral purity requirements are met, although they add loss which can degrade DC efficiency and overall linearity as active devices overcome these losses. At lower microwave frequencies, below 15 or 18 GHz, the filters are generally fabricated on the PCB, but repeatability of this type of filter on a board-to-board basis is generally not acceptable above 23 GHz. This is due to difficulties in operating at or beyond the wet etching process capability used to fabricate accurate coupling gaps that are needed to realize these filters. Endwave has found that if printed filters are incorporated into the PCB, several adverse impacts to cost, performance and on-time delivery can occur as these features become the yield drivers for the PCB manufacturer. This is in direct conflict to the widely held idea that printed features are essentially at zero cost, and must be considered strongly when examining the cost tradeoff between purchasing an additional filter versus printing a “free” one on the PCB.

Figure 3 42 GHz SMT packaged filter response.

Because higher frequency passives in SMT packages are not available through most MMIC suppliers, the need for passive packaged filters previously drove another tradeoff. A choice was made previously to use a wire bond process to allow an alumina filter at the expense of no longer enabling an all-SMT business model. Endwave launched the new SMT filter product to avoid this tradeoff – made possible through the use of the innovative packaging developments. Results of a typical 42 GHz filter are shown in Figure 3 demonstrate the process capability through 45 GHz to reduce spurious, harmonics and RX image rejection while preserving the ability to build an all-SMT transceiver.

These same LO harmonic and TX spurious considerations push voltage-controlled oscillator frequencies higher to minimize LO multiplication factors and the resulting close in hard-to-filter spurious signals. However, the state of the art in low phase noise MMIC VCOs has a high end frequency of approximately 16 GHz, due to the nature of the hetero-junction bipolar transistor process — where the device performance begins to suffer above this range. Based on this, system designers tend to go as high as possible in VCO frequency while remaining under the 16 GHz limit in order to ensure that adequate phase noise is maintained. Moving toward higher frequency VCOs is a clear trade against optimal phase noise that can be obtained through use of a lower VCO frequency with higher LO multiplication factors.

As Figure 1b shows, RX noise figure can be minimized while maximizing RX linearity by protecting the first downconverting mixer from high input power levels. This protection is recommended to be incorporated via the addition of a VVA between the LNA and the mixer. The VVA thus provides a low loss path to the downconverter when the received signal strength is weak. Conversely, the dynamic range of the VVA protects the RX mixer in the case where the received signal is stronger. Note that this tradeoff has been solved through the addition of another tradeoff – where added complexity was incorporated at a slightly higher cost.

At the system level, a strong desire exists to perform system level integration of functions into a single module to reduce cost, while preserving enough isolation between functions to eliminate spurious degradation and instability. This requires a major shift from system complexity to MMIC complexity. Traditional systems have been based on complex systems built from simple MMICs, but today’s trend is toward simpler systems based on more complex MMICs. This has a clearly positive effect on manufacturability of the transceiver module, while putting significant pressure on the MMIC designers.

In any given RF chain, both test yield and margin are directly proportional to the product of variances of the cascaded components (and interconnecting bond wires). Mathematically, the yield of the chain is YT = y1y2…yN, such that limiting the number of MMICs in the chain inherently provides increased overall yield. Since it is also true for the case of addressing variance in performance caused by a shift in any given MMIC, moving to a higher integration level for devices can present some clear advantages.

Figure 4 Simplified transceiver block diagram via integrated MMICs.

Specifically, fewer components in cascade can allow for tolerances on individual devices to be loosened, while having fewer interconnects improves reliability and performance. In addition, an integrated approach like that shown in Figure 4 can enable migration toward a full SMT solution to be far easier than could be done in the block diagram in Figure 1. This is because the high losses would result when adding in 0.4 dB per transition at two or more transitions per MMIC.

In the past, the system design engineer had the freedom to choose single function MMICs from many vendors to gain best in class performance for each slot in the cascade of MMICs. For example, each MMIC function in the block could be from a different vendor, a different process, or both. A PHEMT process upconverter could be chosen with high linearity MESFET-based VVAs, as an example.

However, a key penalty that exists in pushing system complexity to the MMIC level of the supply chain is that grouping multiple functions into a single device forces all those functions to be made by one vendor with one technology (0.15 µm PHEMT low noise process, for example). This is a strong disadvantage from a performance perspective because best-in-class devices from multiple technologies or vendors cannot be chosen when several functions are combined into a single chip – placing an even higher challenge on the MMIC designer.

For this reason, not all functions are likely to be placed on a single chip without dramatic technology development. Figure 4 demonstrates a four MMIC solution which leverages optimal partitioning of functions per technology (note that some systems will use two VCOs for flexibility of frequency planning). In this case, the HBT-based VCOs are fabricated in InGaP, the PHEMT-based integrated converters are fabricated in GaAs, and the power amplifier is fabricated in GaN. A packaged SMT filter is also shown assuming optimal spurious performance is desired. The block diagram, therefore, demonstrates the current state of the art in RF MMIC device integration based on available processes. As mixed process solutions become available, consolidation to fewer MMICs is likely.

Detailed Design Process

In order to design a set of converters that will fulfill the difficult set of tradeoffs listed previously, the design must be approached more like a system design than that of a single function MMIC. System engineers use cascaded analysis tools, which are now gaining popularity with MMIC designers. Cascaded analysis tools are popular, where several critical requirements are calculated simultaneously, including output power, gain, linearity and noise performance as well as variation versus temperature, which allow for rapid optimization.

By applying similar cascaded analysis techniques to the MMIC converters as would normally be done at the system level, significant improvements can be made to yield and performance. Additionally, 2nd spin redesign of the MMICs due to surprises are minimized because there are fewer chances for error. Once an initial cascaded analysis is completed to partition the requirements, the detailed design of the individual pieces can be completed. It should be noted that the SMT interconnects as well as the effect of the plastic overmold must be heavily modeled using 3D simulators and included in the cascaded models as well to properly account for parasitic effects and additional losses.

A key area of focus is gain partitioning between the IF and RF sections. This will ensure the best trade-off between linearity and noise figure (output noise levels) versus dynamic range of the converters as well as spurious implications while maximizing DC efficiency. In addition, choice of mixer architecture as well as on chip LO multiplication factor become crucial. There is little opportunity to filter out undesired LO harmonics that can reach the mixer – as could be done if separate multiplier and converter MMICs were used in a subsystem design.

A secondary consideration is that both LO and image rejection must be suppressed to ensure that these leakage tones do not prematurely degrade linearity of the RF amplifiers that follow the on-chip mixer. Suppression of these tones can be achieved through use of DC feedback injected through the IF inputs of the upconverter. However, in order to achieve the best case performance, the use of SMT-based filters are recommended prior to the power amplifier. Once all individual functions are designed, the detailed specific analysis using linear and nonlinear simulators, as well as 3D and cascaded tools can then be finalized.

Conclusion

A detailed discussion has been presented outlining the business and performance drivers that are the root cause for the fundamental shift in the complexity of RF systems to MMICs design area. In addition, a set of tradeoffs has been presented with solutions enabled by disruptive SMT technology, which is changing the manufacturing paradigm to enable a fully outsourced model. While the case study analyzed was the front-end of a point-to-point radio telecommunications link, the concepts are broadly applicable to several industries that depend on similar technologies.