aging effect, will the phase relation input to output change over time?

These are difficult questions to answer but let me take a stab. I'm attaching a dataset of s-parameters for 10 devices. This will give you some sense of the part to part variations. If you filter the dataset at 3 GHz, you see that the S21 phase varies by around 10 degrees from 21 to 31 degrees. However, these parts probably all come from the same lot. In general GaAs fabrication processes tend to be less well controlled than silicon processes. So if we looked at the same data from a different lot, you might see similar part to part variance but maybe the absolute phases might be different, say, 51 to 61 degrees. 

One other thing to note is that for all of the devices in this dataset, the gate voltages (column G) hav been adjusted so that the devices have a quiescent current of approximately 300 mA (column J). I believe that if you gave all of the devices the same gate voltage, say, exactly -2.6V, you would see more variance in the part-to-part S21 phase. 

Ageing effects are another difficult topic. I've no information on how GaAs devices s-parameters shift over time. I've done a fair amount of work on measuring the ageing effects on silicon devices. What we have seen across a lot of different device classes is that when you turn on a device at time = 0, performance shifts but then settles over a period of approximately 100 hours. After that there is less movement. Here is a short ap note written about long term stability of a voltage reference (AN-713: The Effect of Long-Term Drift on Voltage References | Analog Devices). You will notice in this case that different samples drift in different directions. I can't say if you will get the same behavior from ADPA1106. I've seen cases were all devices drift in the same direction (by different amounts) and also cases where different devices in a batch drift in different directions.

So what do we do with all of this? I would make the following recommendations.

1. If you are building an array, I would recommend ordering all of the devices at the same time. That increases the likelihood that all of the devices in and individual array will be from the same manufacturing lot. 

2. If possible, I would recommend adjusting the gate voltages of each amplifier so that all of the quiescent currents are equal. But, I realize that this may not be practical. 

3. If you are going to calibrate your array, it may make sense, if practical, to burn it in for a few days before calibration

4. If you replace an amplifier in your array, there is a good chance that it will come from another production lot (unless you buy spares when you make your initial purchase). That device may exhibit different absolute S21phase from its older neighbors. Also, its phase may shift a little over the first 100 hours or so. Back to point 2, I think that adjusting the gate voltage of that new devices so that it has the same target quiescent current as the rest of the devices in the array, has value. 

Eamon

 XLSX