mmWave RF Driver Amplifiers with Better Engineered Tradeoffs

Challenge: Reaching stability with minimized prototyping

The origin story of the ADM-8007PSM is a tale of the importance of design for context (DfC) in complex RF systems. Customers asking Marki for help designing a driver amplifier for its mixers is no accident.Before DfC, RF system designers had the chore of choosing components for a spot in the signal chain based on generic data sheet parameters selected by the manufacturer. System-level simulations often sampled a few frequency points, hoping no interpolative surprises lurked between points.

Higher frequencies, wider bandwidths, and complex modulation shattered that thinking. The central theme of DfC is models and simulations need to reflect how RF components work in an application –exposing interactions between components, packaging, circuit board layouts, and cross-domain effects.

“Customers have significant input to requirements, but most of our products are specified and developed by the engineers who will make the tradeoffs that look the best,” says Doug Jorgesen, VP of Systems and Applications at Marki Microwave. The objective for the ADM-8007PSM was designing a driver amplifier that would deliver optimum performance from the local oscillator (LO) amplifier-mixer combination, shown in Figure 1, placed in a typical mmWave signal chain.

“The biggest question mark on the ADM-8007PSM was doing this two-stage design to hit the gain target at high frequency across a wide bandwidth in an unfamiliar semiconductor process in a package with unknown bias circuits, and have it all end up stable,” continues Jorgesen. High power and efficiency weren’t primary concerns; Jorgesen says getting more than 50% efficient on a super-broadband design (more than an octave wide) is hard. He also says Marki does not do extensive load-pull analysis in most cases, partly because the intended driver amplifier load is known – a Marki mixer – and partly because efficiency isn’t an optimization criterion for the design.

In their prior workflow, Marki engineers would use their experience to guess the most sensitive parts of adriver amplifier, create perhaps four or five or more candidate designs with varying parameters, build them all out and take physical measurements, and pick the one they liked best to proceed. “If our guesses were correct, and parts didn’t oscillate due to layout or bias issues, we’d still end up with up to three spins to reach a finished amplifier product,” Jorgesen observes.

What Jorgesen describes is an ideal use case for circuit-EM co-simulation using ADS, doing the necessary parameter exploration virtually and improving the chances of first-pass success at fabrication.

Solution: Modeling parasitics, bias, and packaging

Consider a non-trivial point of concern in simulations matching real-world measurements for an advanced high-frequency semiconductor process with complex non-linearities: the behavioral models must be highly detailed. Marki Microwave brings extensive modeling experience to bear on the opportunity.

“Foundry process design kits (PDKs) are a starting point for us,” explains Jorgesen. “We’ll take engineering samples from the foundry, measure S-parameters from those very carefully, and use those results to either modify the PDK they provide or develop a standalone model for our needs.”

The next step is embedding the robust S-parameter measurements within Marki’s passive structure modeling. More from Jorgesen: “It’s pretty hard to make a non-linear transistor model, but it’s relatively easy to make a linear transition model taking the measured S-parameters and fitting those in.” Marki also uses a few hacks to improve parasitic representations in the structure, like adding bias-variable capacitance or inductance, or even negative capacitance, which most foundries say is a no-no in PDKs.

Marki also invested considerable effort in choosing the framework for the transistor models. “We went pretty deep with the Keysight transistor team, David Root and his people, to understand the best transistor model for us to use,” Jorgesen shares. Those discussions led Marki’s teams to the EEHEMT transistor models in ADS. “We ended up with the S-parameter embedding, and we can then play with other elements to make scalable models for studying specific aspects of the transistors in the amplifier.” Some moreelements Marki uses for its model refinements and enhanced characterization:

• Noise figure profiles from straightforward noise measurements.

• Self-heating and current change under drive, secondary effects requiring attention.

• Bias voltages varied across simulation scenarios for key performance metrics.