V
Viettel Semiconductor
Guest
If you've spent time on an X-band T/R module, you know the marketing slide: "single-die front-end, PA + LNA + switch, fully integrated." It reads like a simple parts-count win. In practice, it's one of the harder integration problems in GaN MMIC design — and the reason so few vendors ship a genuinely production-qualified single-die X-band FEM.
Isolation, not gain, is the bottleneck. The PA in transmit mode is putting several watts through a path physically inches — sometimes microns — from an LNA built to detect signals at the noise floor. Any leakage through the switch, through substrate coupling, or through the shared bias network desensitizes the receive chain or, worse, damages the LNA front end during transmit. Getting 40+ dB of isolation on a shared die, without a mechanical relay or an off-chip switch to lean on, forces trade-offs in switch topology (series-shunt vs. traveling-wave) that cost insertion loss and linearity elsewhere.
The switch is not a passive afterthought. A GaN switch sitting between a multi-watt PA and an antenna port has to handle real power without generating harmonics that show up in spec, without adding meaningful insertion loss ahead of the LNA in receive mode, and without a channel resistance that varies with the PA's thermal state. Insertion loss ahead of an LNA hits noise figure directly — every 0.3 dB of switch loss is 0.3 dB of NF you can't design out downstream.
Thermal coupling is a system problem hiding in a device problem. The PA runs hot. The LNA needs a stable bias point to hold its noise figure and gain flat across temperature. On a single die, the LNA is sitting in the PA's thermal gradient, not in a controlled ambient. Bias circuits that looked stable in simulation at 25°C can drift once the PA has been keyed for a few hundred milliseconds next door. This is the failure mode that doesn't show up until you have real silicon under real duty cycles — which is exactly why so many "integrated FEM" datasheets stop at post-layout simulation and never publish measured PA-on/LNA-noise-figure data.
Process yield compounds every block's risk. A PA optimized for power and efficiency wants different bias and layout margins than an LNA optimized for noise figure, and both differ from what a low-loss switch wants. Putting all three on one die means a defect or process corner that only marginally affects the PA can kill the LNA's noise figure spec, and now your yield is the product of three sub-yields instead of the yield of your best block. This is the quiet reason single-die FEMs are harder to get to volume than three separate, best-of-breed die in a module.
That's the bar we designed VPF1010, our X-band GaN-on-SiC FEM, against: PA, LNA, and switch integrated on a single die, characterized and silicon-proven rather than simulated. We'd rather talk through the actual isolation and NF-under-drive numbers with the engineers evaluating it than lead with a features list.
If you're working on X-band T/R module architecture and want to compare notes on where single-die integration helps vs. where a discrete approach still wins for your application, the VPF1010 datasheet and technical contact are on our [Design&Reuse listing] — happy to go deep on the measured data.
The problem looks like layout. It isn't.
On paper, combining a power amplifier, a low-noise amplifier, and a T/R switch on one GaN-on-SiC die is a floorplanning exercise: put the blocks where they fit, route the RF paths, done. The actual constraints are electrical, thermal, and process-driven all at once, and they fight each other.Isolation, not gain, is the bottleneck. The PA in transmit mode is putting several watts through a path physically inches — sometimes microns — from an LNA built to detect signals at the noise floor. Any leakage through the switch, through substrate coupling, or through the shared bias network desensitizes the receive chain or, worse, damages the LNA front end during transmit. Getting 40+ dB of isolation on a shared die, without a mechanical relay or an off-chip switch to lean on, forces trade-offs in switch topology (series-shunt vs. traveling-wave) that cost insertion loss and linearity elsewhere.
The switch is not a passive afterthought. A GaN switch sitting between a multi-watt PA and an antenna port has to handle real power without generating harmonics that show up in spec, without adding meaningful insertion loss ahead of the LNA in receive mode, and without a channel resistance that varies with the PA's thermal state. Insertion loss ahead of an LNA hits noise figure directly — every 0.3 dB of switch loss is 0.3 dB of NF you can't design out downstream.
Thermal coupling is a system problem hiding in a device problem. The PA runs hot. The LNA needs a stable bias point to hold its noise figure and gain flat across temperature. On a single die, the LNA is sitting in the PA's thermal gradient, not in a controlled ambient. Bias circuits that looked stable in simulation at 25°C can drift once the PA has been keyed for a few hundred milliseconds next door. This is the failure mode that doesn't show up until you have real silicon under real duty cycles — which is exactly why so many "integrated FEM" datasheets stop at post-layout simulation and never publish measured PA-on/LNA-noise-figure data.
Process yield compounds every block's risk. A PA optimized for power and efficiency wants different bias and layout margins than an LNA optimized for noise figure, and both differ from what a low-loss switch wants. Putting all three on one die means a defect or process corner that only marginally affects the PA can kill the LNA's noise figure spec, and now your yield is the product of three sub-yields instead of the yield of your best block. This is the quiet reason single-die FEMs are harder to get to volume than three separate, best-of-breed die in a module.
Why this matters for radar and Satcom integrators
For AESA T/R module designers, this isn't an academic point. Every one of these trade-offs shows up as a line item you have to test for at the module level: TX-to-RX isolation under full PA drive, RX noise figure with the PA keyed nearby on the same panel, switch insertion loss vs. temperature, and long-term reliability under duty cycle. A vendor that can hand you a single-die FEM with measured, silicon-proven data for all of these — not simulated, not "expected" — has done the hard part of your qualification work before you ever get a sample in hand.That's the bar we designed VPF1010, our X-band GaN-on-SiC FEM, against: PA, LNA, and switch integrated on a single die, characterized and silicon-proven rather than simulated. We'd rather talk through the actual isolation and NF-under-drive numbers with the engineers evaluating it than lead with a features list.
If you're working on X-band T/R module architecture and want to compare notes on where single-die integration helps vs. where a discrete approach still wins for your application, the VPF1010 datasheet and technical contact are on our [Design&Reuse listing] — happy to go deep on the measured data.