Optimization of sb-heterostructure diode for low noise detection

Current millimeter wave imaging cameras based on square-law detector diodes require an RF low noise amplifier (LNA) to boost the signal above the detector noise floor. Sb-heterostructure diodes, fabricated from epitaxial layers of InAs and AlGaSb, have shown very high zero-biased sensitivities at W-band, but the devices studied were not optimized for low noise [1-3]. Zero-biased devices have an important a priori advantage over the commonly used biased Schottky diode detector because removing the bias eliminates 1/fnoise and provides the potential for ultra-low noise performance without the need for pre-amplification. The remaining noise is predominantly Johnson noise, Sv =4kTRj (per Hz), where RJ is the junction resistance. A key figure of merit is the Noise Equivalent Power (NEP), a measure of the minimum detectable power, given by SV"I2/S, where S is the sensitivity in volts per watt. To lower the noise, RJ can be decreased from the -1OKQ value of the previous diodes. The basic diode active region contains highly doped n-type InAs, an AlSb barrier, an un-doped AlGaSb layer, followed by a highly doped GaSb (p) to InAs (n) tunnel junction [1-3]. The key layer being altered here is the AlSb layer. It controls the overall current flow through the device and thus the junction resistance. We have thinned it from its previous values of 32-39A to 1 5-20A to lower the junction resistance. The I-V curve of a typical 2x2 [m2 diode with a 15iA AlSb barrier is shown in Fig. 1. A polynomial fit to the important figure of merit y, the I-V curvature divided by the slope [4], is found to be 23V'. This is somewhat less than the ~40V-1 curvature found with the thicker AlSb barriers. The diode with a 20A AlSb produced an intermediate curvature of 27V-', illustrating the expected trend of approaching ohmic behavior as the barrier is thinned. Sparameters measured to 40 GHz are used to extract a standard equivalent circuit model. Figure 2 shows a table ofthe extracted results for the diodes with 15A and 20A barrier thicknesses. The values differ by about 10-15% across a wafer due to variations in epitaxy, processing tolerances, and measurement uncertainties. These values, along with the polynomial, are utilized in a user defined non-linear model in Microwave Offi1ce [5] with the results being displayed in Fig. 3. This result is similar to the measured data in Fig. 4. for which the diode was placed in a housing with 5 mil alumina circuits and tuned for optimum sensitivity. In both cases, the maximum sensitivity is close to 2500 V/W. An open X/4 line at 36 GHz was added to reflect the RF signal back into the diode to offset the effects from capacitor loss on the DC output side. As the line acts as a narrowband bandstop filter, the bandwidth of the circuit was reduced. Associated measured low frequency noise of the diode is shown in Fig. 5. The best fit is to a Johnson noise power equivalent to that of a 630Q resistor, quite close to the junction resistance plus -38Q of series resistance. This shows that under the zero-bias condition the noise follows the simple behavior predicted. Using this noise power and the peak measured sensitivity of2700V/W in the equation for NEP yields 1.2 pW/Hz"2. We believe that our Rs values can be reduced considerably as SEM photographs revealed significant undercutting. This plus further shrinking the diode area and capacitance should lead to NEP values well under 1 pW/Hz12.