We show photographs of the inside and outside of the 4--16 GHz MMIC LNA chassis including the matching network in Figure 1. When aiming for lowest noise performance, the matching network and performance of the first stage ultimately sets the noise performance. To avoid substrate losses, and consequently degraded noise performance, an external input matching network on a low loss substrate was used. To decrease the magnitude of S 11 (input reflection coefficient) and improve stability, a high inductance source microstrip was used on the first transistor.To achieve flat in--band noise, an impedance matching tradeoff has to be done. The minimum noise temperature of an InP HEMT varies nearly linearly with frequency [1]. Therefore, the best approach to obtain a flat noise temperature within a certain frequency band is by noise matching the first stage at the upper frequency limit, while allowing a certain mismatch at the lower frequencies. The second and third stages were matched for flat gain, stability and output match. The LNA has a common bias network for all three stages.
S2: Additional information on noise model and parameter extractionThe noise model and method used for noise parameter extraction is the same as in [2]. The principle of the model is that the noise in the device can be represented by thermal and non--thermal sources. Thermal noise is due to resistive elements in the amplifier, with the magnitude depending on the value of the resistances and the physical temperature. The only non--thermal noise source is hot--electron noise source in the output of the InP high electron mobility transistors (HEMTs). This noise source is typically modeled as an elevated drain temperature T d that is a few orders of magnitude larger than the physical temperature of the HEMT. If the ambient temperature T a of the LNA is known and the noise parameters are measured using the Agilent Noise Figure Analyzer, T d can be obtained by fitting the simulated and