Inferensys

Glossary

Power-Added Efficiency (PAE)

Power-Added Efficiency (PAE) is a metric quantifying a power amplifier's ability to convert DC supply power into added RF output power, critical for thermal and energy budgets.
Supply chain manager using AI negotiator on laptop, supplier data visible, casual office afternoon setup.
RF METRIC

What is Power-Added Efficiency (PAE)?

Power-Added Efficiency (PAE) is the critical figure of merit that quantifies a power amplifier's effectiveness in converting DC supply power into useful RF output power, accounting for the RF input drive.

Power-Added Efficiency (PAE) is defined as the ratio of the added RF power (output minus input) to the total DC power consumed by the amplifier. It is calculated as PAE = (P_RF_out - P_RF_in) / P_DC. Unlike drain efficiency, PAE accounts for the RF input drive power, providing a true measure of net power gain relative to energy cost.

High PAE is paramount in mmWave phased arrays and mobile handsets to minimize thermal dissipation and extend battery life. The metric directly influences the thermal budget and linearity trade-off; amplifiers often operate in back-off from peak PAE to meet stringent ACLR requirements, a balance that digital predistortion helps optimize by enabling operation closer to the efficiency peak.

EFFICIENCY DETERMINANTS

Key Factors Influencing PAE

Power-Added Efficiency is not a fixed device parameter but a dynamic metric governed by semiconductor physics, circuit topology, and operating conditions. Understanding these factors is critical for optimizing energy budgets in mmWave phased arrays.

01

Semiconductor Material Properties

The choice of substrate fundamentally limits achievable PAE. Gallium Nitride (GaN) offers superior efficiency at mmWave frequencies due to high electron mobility and wide bandgap, enabling higher power density with less waste heat compared to Gallium Arsenide (GaAs) or Silicon CMOS.

  • GaN-on-SiC achieves PAE > 40% at 28 GHz
  • CMOS typically peaks at 15-25% in mmWave bands
  • Material choice directly impacts thermal memory effects
> 40%
GaN PAE at 28 GHz
02

Amplifier Class of Operation

The conduction angle defines the theoretical efficiency ceiling. Class-A amplifiers conduct for 360° with a maximum theoretical PAE of 50%, while Class-AB balances linearity and efficiency. Doherty architectures use load modulation to maintain high PAE at back-off.

  • Class-A: 50% max theoretical, poor average efficiency
  • Class-AB: Common compromise for linearity vs. efficiency
  • Doherty: Maintains PAE at 6-10 dB Output Back-Off (OBO)
50%
Class-A Theoretical Max
03

Peak-to-Average Power Ratio (PAPR)

Modern wideband signals like OFDM exhibit high PAPR, forcing the PA to operate at significant back-off from its peak-efficiency point. A signal with 10 dB PAPR may force a PA to operate at 10 dB below saturation, drastically reducing average PAE.

  • 5G NR signals: 8-12 dB PAPR typical
  • Crest Factor Reduction (CFR) can improve PAE by 3-5 percentage points
  • Envelope tracking recovers efficiency lost to back-off
8-12 dB
5G NR Signal PAPR
04

Load Impedance and Matching Networks

PAE is highly sensitive to the impedance presented at the transistor's current generator plane. Load-pull analysis identifies optimal impedance contours for maximum PAE. In phased arrays, active impedance mismatch during beam-steering pulls the PA away from this optimum.

  • Optimal Z_load varies with frequency and power level
  • Beam-steering causes channel-specific PAE degradation
  • Antenna crosstalk further distorts load impedance
5-15%
PAE Drop from Mismatch
05

Operating Frequency and Bandwidth

PAE inherently degrades as carrier frequency increases toward the device's f_T and f_max limits. At mmWave frequencies, parasitic capacitances and reduced gain compress efficiency. Wide instantaneous bandwidths introduce frequency-dependent memory effects that further erode PAE.

  • GaN f_max > 200 GHz enables viable mmWave PAE
  • Bandwidths > 400 MHz challenge matching network design
  • Trapping effects in GaN become more pronounced at high frequencies
> 200 GHz
GaN f_max
06

Thermal Management and Self-Heating

Junction temperature rise directly degrades carrier mobility and PAE, creating a positive feedback loop where reduced efficiency generates more heat. Thermal memory effects cause slow PAE fluctuations tied to signal envelope history.

  • Every 10°C rise can reduce PAE by 1-2 percentage points
  • GaN-on-Diamond substrates offer 3x better heat spreading
  • Envelope tracking reduces DC power dissipation and thermal load
1-2%
PAE Loss per 10°C Rise
POWER-ADDED EFFICIENCY

Frequently Asked Questions

Clear, technically precise answers to the most common questions about Power-Added Efficiency (PAE), its calculation, and its critical role in power amplifier design and system-level thermal management.

Power-Added Efficiency (PAE) is a critical figure of merit that quantifies a power amplifier's ability to convert DC supply power into useful added RF output power, accounting for the RF input drive power. It is calculated as PAE = (Pout_RF - Pin_RF) / Pdc_DC, where Pout_RF is the RF output power, Pin_RF is the RF input power, and Pdc_DC is the total DC power consumed from the supply. This metric is distinct from drain efficiency because it subtracts the input signal power, providing a more accurate measure of the amplifier's true power conversion effectiveness. A high PAE is essential for minimizing heat dissipation and extending battery life in mobile devices.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.