Eleven years ago I had the opportunity to start designing RF/microwave amplifiers and I also had MultiMatch at my disposal. Nevertheless I designed my first amplifier (0.5-2GHz LNA) using only a general simulation software. Then I had to design a 2-6GHz one. At the same time I was getting to know MultiMatch and … that was that. I was hooked. The amplifier was synthesized using the MultiMatch design procedures and then its schematic was translated to a general type simulator to check and polish the performance with its more accurate discontinuity models. No optimization was used for that but only some simulation tuning. Then came a 2-18GHz 5 stage LNA. Both designs were first-time-right. No other iterations were necessary. What was even better was that the amps didn’t need tuning in regular production.

The same design results continued to be the case amplifier after amplifier for the more than 100 amplifiers I have brought to production. There were only 5-6 cases where I had to do a second design iteration but this has nothing to do with MultiMatch. It was the result of my own laziness to not properly model some discontinuities or parasitics.

Some of the types and applications of these amplifiers are as follows: broadband up to 18GHz LNAs and high-dynamic range amps for electronic warfare; high-dynamic range and up to 30W Class A and up to 100W Class AB amps for the mobile phone infrastructure; LNAs and medium power (4W) amps for the satellite communications (2.5-2.7GHz, 3.6-4.2GHz); 50-500MHz LNAs, medium and high power up to 100W amps for the two-way mobile communications.

A common characteristic of the amplifiers designed by using MultiMatch is that they are relatively broadband, let’s say 350-500MHz, 750-1000MHz or 1.7-2GHz, but will have the performance of the typical narrowband amplifiers that are on the market, i.e., same NF, P1dB, Psat and efficiency, but for much broader bandwidths. When very broadband amplifiers are designed they will outperform the industry standard broadband ones on every parameter.

Today I cannot possibly imagine designing amplifiers without using MultiMatch. It will be torture if it comes to it and I’ll most probably give up engineering.  This may sound as a strong statement but I am not one of those engineers who likes to hack out quickly not well thought designs, usually still using only pencil and the Smith Chart, and then bleed to death at the test bench for many days and nights, and when the product works half-well to throw it in production creating a nightmare there. These engineers proudly think of themselves as hackers but I’ve never been mystified or fascinated by their false feeling of self-importance. The way I like to develop amplifiers is to thoroughly design them using intelligent software design tools, to pay meticulous attention to the details and to properly model every component and element of the design. Then when the first prototype of the designed amplifier is built and you switch the power supply on, the performance is spot on and exactly the way you wanted it. Then the same must be happening in production, amplifier after amplifier. That’s how I like it and that’s how it works when amplifiers are designed with MultiMatch.

Strangely enough, most of the companies in the RF/microwave amplifier industry, though not the MMIC producers, are still developing their amplifiers avoiding the use of RF/microwave design software. The trial-and-error method of design is a deeply entrenched culture. I have to recognize that there are people that achieve some ingenious designs by the trial-and-error method. But does this method provide optimum performance and trouble free amplifiers in production, and all of this in a very short design cycle?

For the MMIC designers it is cost prohibitive to use the trial-and-error method to physically produce wafer after wafer with incremental changes. But they still do trial-and-error searches by using general simulator/optimizer software such as ADS, Microwave Office, etc. The issue here is how the designer chooses the initial schematic and if it will provide optimum performance. Good initial schematics are usually achieved only after years of experience. Let’s say we somehow have a schematic that shows good performance in simulations. Now we have to do a yield analysis and possibly yield optimization and the latter will take considerable computer resources. What happens when we can’t achieve the desired yield because the initially chosen circuit is too tolerance sensitive? We have to start from scratch. The trial-and-error method by using simulation and optimization can put you into a time consuming design iterations loop.

The design philosophy of MultiMatch is based on fundamentally different (kind of a reverse) to the “common sense and culture” logic. The first step of the MultiMatch design approach is the synthesis of device-modification networks for each stage of the amplifier. The device-modification networks are feedback and loading networks (typically each one containing a resistor in combination with a reactive component) which are placed around and close to each transistor. It is for the first time in MultiMatch that these networks are grouped together under a common name because of their similar functions. They are also synthesized together in combinations of two at a time. The device-modification capabilities are used to remove inherent gain slopes, to stabilize the transistors, to reduce the gain-bandwidth constraints associated with the input and output impedances of the transistors used, to move the points of optimum NF or optimum P1dB close to the maximum gain points and last but not least, to de-sensitize the transistor behavior toward the component tolerances. Then follows (if necessary) the synthesis of lossless impedance-matching networks which is based on real-frequency numerical systematic searches using the transformation Qs as search parameters. Here the tolerance sensitivity is addressed again first because MultiMatch is trying to keep the transformation Qs as low as possible and because it presents, as a result of the synthesis, multiple solutions which show simultaneously the performance error and the tolerance sensitivity. The solution chosen could be the one for best performance, but if the tolerance sensitivity is considered too high, a solution with compromise between performance and tolerance sensitivity could be chosen. When a solution with high tolerance sensitivity has to be chosen, the transformation Qs show which components tolerances should be controlled tighter. The design approach philosophy of MultiMatch guarantees solutions that provide simultaneous optimum performance and low tolerance sensitivity.

On top of all the fundamentally different features come the power parameters. They add the liberating capability to design for P1dB when the only available data are the S-parameters, the biasing point and the voltage and current limitations of the transistors. The S-parameters cannot give any information about P1dB but MultiMatch first allows a linear model to be extracted from the S-parameters of the particular transistor used and then, using this linear model, derives the power parameters. The power parameters originate in the load-line approach applied at RF and microwave frequencies in which it is assumed that the voltage and current across the intrinsic generator of the linear model behave linearly till the point of hard clipping. This point is taken to be the 1dB compression point of the gain and therefore serves to estimate P1dB. This is also what Steve Cripps is showing in his approach, but his approach is limited by the simplistic equivalent model and clipping region used. If the transistor model is more complex and external feedback and/or loading resistors are used, his approach can not help much. The power parameters, elegantly derived by Pieter Abrie using linear mapping functions, overcome any complexity in the transistor model or the networks used in the amplifier. By using the power parameters and the clipping region specified for each transistor, MultiMatch predicts the maximum P1dB of the transistors, derives the load-pull contours, synthesizes the load for any desired P1dB and when the amplifier full schematic has been synthesized, simulation and optimization (if necessary) of P1dB can be performed simultaneously for all the stages.

The MultiMatch Amplifier Design Wizard is a specialized software tool. It is the specialized tools that nowadays allow circuits and devices with optimum performance to be designed. The general simulation/optimization software programmes should be used at the final stage of the design to check and polish the performance using their higher accuracy of modelling and simulation of the components and devices.

Ivan Boshnakov