Self-Heating Effect

The transient thermal response of the transistor's channel to dissipated power is characterized by a thermal impedance (Zth) profile. Unlike a static thermal resistance, Zth captures the time-dependent heating and cooling dynamics. This impedance is typically modeled using a multi-stage RC ladder network (Foster or Cauer model), where each RC stage represents a different thermal time constant corresponding to heat propagation through the die, die-attach, and package layers. These time constants range from microseconds to milliseconds, directly influencing the memory bandwidth of the self-heating effect.

As the channel temperature rises due to instantaneous power dissipation, the electron mobility in the semiconductor material decreases. This results in a transient reduction in transistor gain that lags behind the RF envelope. Key characteristics include:

• Mobility degradation: Higher lattice temperatures increase phonon scattering, reducing carrier velocity.
• Envelope-dependent sag: The gain compression is not instantaneous but follows the thermal time constant, creating a hysteresis-like distortion pattern.
• Signal history dependence: The gain at any instant depends on the average power envelope over the preceding thermal memory window, violating static AM-AM assumptions.

Self-heating induces a transient phase modulation that is distinct from AM-PM distortion. The mechanism involves:

• Temperature-dependent capacitances: The gate-source (Cgs) and gate-drain (Cgd) capacitances shift with temperature, altering the device's input and feedback impedance.
• Threshold voltage drift: The transistor's threshold voltage (Vth) has a negative temperature coefficient, causing a dynamic shift in the conduction angle and operating point.
• Propagation delay variation: Changes in the effective electrical length of the transistor's intrinsic elements modulate the phase of the amplified signal. This phase shift is a function of the thermal envelope history, not just the instantaneous input amplitude.

The self-heating effect exhibits a strong dependence on the envelope frequency of the modulated signal. This creates a frequency-selective memory response:

• Low envelope frequencies (< 1 MHz): The channel temperature can track the instantaneous power envelope closely, leading to significant gain and phase modulation that follows the signal dynamics.
• High envelope frequencies (> 10 MHz): The thermal capacitance of the device filters out rapid power variations, causing the channel temperature to stabilize at an average value. The self-heating effect transitions from a dynamic distortion to a quasi-static operating point shift.
• Spectral shaping: This frequency-dependent behavior shapes the distortion spectrum, contributing to asymmetric intermodulation products and complicating wideband linearization.

In GaN HEMT devices, self-heating is coupled with charge trapping effects, creating a complex electro-thermal memory. The interaction manifests as:

• Gate lag: Surface traps modulate the effective gate potential with slow time constants (microseconds to seconds), interacting with the thermal transient to produce a combined gain sag.
• Drain lag: Buffer traps in the carbon-doped GaN buffer layer respond to the drain voltage swing, and their time constants are themselves temperature-dependent, creating a nonlinear coupling between thermal and trapping dynamics.
• Knee voltage walkout: The combined effect of increased temperature and trapped charge increases the knee voltage, reducing the available voltage swing and efficiency under dynamic operation.

Self-heating is classified as a long-term memory effect in power amplifier behavioral modeling, distinguishing it from short-term electrical memory effects:

• Long-term memory: Effects with time constants comparable to or longer than the inverse of the signal bandwidth. Self-heating falls here because thermal time constants (microseconds to milliseconds) are often longer than the symbol period of wideband signals.
• Separation from electrical memory: Bias circuit impedance and matching network dispersion cause short-term memory (nanoseconds), while self-heating and trap effects dominate the long-term memory response.
• Modeling requirement: Accurate digital predistortion for wideband signals requires models that capture both short-term and long-term memory, such as Volterra series with thermal kernels or augmented memory polynomial models with envelope-dependent thermal terms.