We report the use of diffuse reflectance spectroscopy for active, closed-loop control of substrate temperature during the growth of a modulation doped heterostructure. Measurement and control of substrate temperature is a common difficulty for molecular beam epitaxy (MBE) growth of semiconductor structures. Conventional MBE employs a thermocouple (TC) in the vacuum gap between the heater coils and the substrate. In steady-state, the temperature offset between the TC and the sample surface can be of the order of 100 °C, but can be calibrated with pyrometry or by observing known surface changes with reflection high energy electron diffraction. Even under calibrated conditions, the TC reading can significantly lag the actual substrate surface temperature during transients, resulting in heterolayer deposition well above or well below growth temperatures. Temperature errors can lead to heterolayer changes such as segregation, desorption, and changes in dopant activation. Diffuse reflectance spectroscopy (DRS) has been used to monitor the absorption edge of the semiconductor substrate, which can be correlated to the temperature.
Because it is a direct measure of a physical property of the bulk substrate, DRS does not suffer from the time lag during transients that thermocouples do. We have exploited this capability to sample and control the substrate temperature with the DRS and substrate heater in active, closed-loop control. To examine the effect of the temperature lag experienced during conventional MBE, we have grown identical pairs of GaAs / InxGa1-xAs / AlyGa1-yAs pseudomorphic high electron mobility transistors (pHEMTs). For one pHEMT in each pair, the input signal for substrate temperature control was the TC; for the other, it was the DRS. Under TC control, an overshoot of up to 70 °C was observed during the temperature upramp following the lower-temperature deposition of the InxGa1-xAs layer. This overshoot was eliminated under DRS control. The DRS controlled sample exhibits a strong peak in gate-drain conductance at 1 V depletion; this peak is absent in the TC controlled sample. We will discuss the implications of this difference for pHEMT growth procedures and for device performance.