Table des matières
In Ka-band satellite communication and millimeter-wave radar transmitters, GaN MMIC power amplifiers have become the only viable choice — no other semiconductor can deliver multiple watts in just a few square millimeters.
But there’s a price: heat.
Several watts of heat concentrated on a chip smaller than a grain of rice. Junction temperature shoots past limits in seconds.
So the real engineering problem never changes: raise PAE, control thermal resistance.
No fluff. Here’s what actually works at 30 GHz and above: why GaN gets hot and inefficient, how to fix it in circuitry, how to contain it in packaging, and which specs actually matter.
1. First, Understand Why GaN Runs Hot and Inefficient at mmWave
Many engineers use GaN just fine at lower frequencies. Then they move above 30 GHz and something feels wrong — same process, lower efficiency, double the heat.
Two main culprits:
1.1 Output impedance drops drastically — matching becomes a brute-force problem
At mmWave, parasitic capacitance and on-resistance effects amplify significantly.
The result: very low output impedance.
Matching to a 50Ω system requires a high transformation ratio. Microstrip loss skyrockets — and that loss directly eats into PAE.
A simple way to think about it: At low frequencies, matching feels like pushing a door. At mmWave, it’s like pushing a wall in mud. Loss is inevitable.
1.2 Heat doesn’t spread — it concentrates into “hot spots”
GaN commonly uses SiC or diamond substrates. The substrate itself conducts heat reasonably well.
But under high-frequency, large-signal operation, heat concentrates right at the drain channel — forming micron-level hot spots.
Good substrate thermal conductivity doesn’t help if heat can’t escape that tiny spot fast enough.
This is called thermal localization — the real challenge in mmWave GaN thermal design.
2. Circuit-Level Fixes: Two Topologies That Actually Help
On-chip architecture sets the upper limit for PAE. At mmWave, these two approaches are proven in real engineering.
2.1 Continuous Class-F / Class-J — Stop Chasing High-Order Harmonics
Traditional switched-mode PAs reduce power dissipation by precisely controlling harmonic impedance to minimize voltage-current overlap.
But at mmWave, precisely controlling third-order and higher harmonics is nearly impossible — parasitics are too sensitive.
The practical engineering solution: continuous Class-F or Class-J.
You allow harmonic impedance to have some tolerance (either capacitive or inductive) on the Smith chart.
You sacrifice a small amount of theoretical peak efficiency, but in return you get stable high PAE over a wide bandwidth — with an order of magnitude lower implementation difficulty.
Bottom line: Don’t chase textbook harmonic control at mmWave. Continuous-mode topologies are what actually work in production.
2.2 Power Combining — Insertion Loss Is the Real Metric
Single devices don’t deliver enough power. You have to combine multiple cells. But combining comes with a cost: insertion loss.
- Traditional Wilkinson combiners: High loss at mmWave — not recommended
- Low-loss spatial combining or waveguide-based combining: Keep combining loss under 0.5 dB
0.5 dB doesn’t sound like much, but in a transmit chain, that’s 10–15% of your output power wasted.
Conseil de sélection : Many vendors don’t explicitly spec combining network loss. But you can back-calculate it from the gap between individual device capability and total PAE.
3. Thermal Management: Two Layers, Both Critical
Thermal design can’t rely on packaging alone. At mmWave, the heat path looks like this:
GaN channel (hot spot)
↓
Diamond heat spreader (if used)
↓
SiC / Diamond substrate
↓
AuSn eutectic solder
↓
Package (CuMoCu or other heat sink)
Every single layer matters.
3.1 GaN-on-Diamond — Cut Thermal Resistance at the Source
Diamond has thermal conductivity of ~2000 W/m·K — over 4x higher than SiC.
Growing GaN directly on diamond substrate reduces channel-to-substrate thermal resistance by 40% or more.
This is the most effective source-level solution. It’s expensive, but for high-power CW applications, there’s no alternative.
3.2 Eutectic Bonding — Void Rate Is the Devil in the Details
Between the chip backside and package, the industry standard is Au80Sn20 gold-tin eutectic.
Good thermal conductivity. Strong adhesion.
But the most common real-world failure isn’t the material — it’s void rate.
Microscopic bubbles create localized thermal resistance. One spot on the chip runs hot. Then it burns.
Hard requirement: Void rate < 5%. If a vendor doesn’t offer X-ray inspection data during acceptance, that’s a red flag.
4. Four Specs That Actually Matter (Not Just Output Power)
Datasheets are full of impressive numbers. But these four determine whether your system actually works:
| Paramètre | Industry Benchmark (Ka/Q-band) | Why It Matters |
|---|---|---|
| Saturated output power (Psat) | ≥ 40 dBm (10W) | Sets the upper limit of radar detection range |
| Power-added efficiency (PAE) | 25%–35% | Every 5% improvement shrinks power supply and cooling volume by nearly 1/3 |
| Thermal resistance (Rth) | ≤ 4.5 °C/W | Determines junction temperature and lifetime under CW operation |
| Package void rate | < 5% (AuSn eutectic) | Not on the datasheet — but this is what actually kills chips |
Heads-up on PAE: Always ask for PAE over your actual operating bandwidth, not the peak value at a single frequency. Peak PAE is usually narrow and meaningless in real systems.
5. Une conclusion sincère
A good mmWave Amplificateur de puissance au GaN est not about who has the highest output power. It’s about the balance between PAE and thermal resistance — because that balance determines whether your system can actually be deployed.
- If PAE is 5% worse, your heatsink doubles in size and your power supply gains two pounds — and your program manager will not be happy.
- If thermal resistance isn’t controlled, after 1000 hours of burn-in, output power drops 3 dB — and your reliability is gone.
When evaluating, ask these three questions:
- “Is your PAE measured under CW or pulsed conditions?”
- “Is thermal resistance junction-to-case or junction-to-ambient?”
- “Do you provide X-ray void rate inspection reports for eutectic bonding?”
Ask these, and you’ll filter out half the “impressive datasheet” products on the market.
