RFIC ESD diode that steals your NF budget before the first stage ever sees the antenna

Started by LunaWei · 2026-11-13 01:46 UTC · 39 views · 5 replies
L Admin
[Technical article / FAQ, rfforum Editorial Team]

An RFIC input that meets ESD on the datasheet can still burn 0.3–0.8 dB of noise figure in diodes and clamps the app note never puts in the NF cascade. The first LNA stage never gets the antenna match you simulated; the ESD network sits in series with that match and owns part of the budget.

1. What the clamp actually does at RF
- ESD diodes add shunt capacitance and lossy resistance at the pin. At a few GHz that capacitance is already a matching element; at tens of GHz it can dominate the input Q.
- A "robust" dual-diode + series resistor path that passes HBM can look like a few tenths of a dB IL plus a shifted Gamma_opt. Your discrete matching network then fights a different load than the bare die model.
- Package bond wires and the ESD cell are often one lump in the S-parameters; stripping ESD "for RF performance" without a re-qualified package is not a free lunch.

2. Freeze list before you blame the LNA core NF
- Whether the S-parameters and NF data include the ESD cell (and which package)
- De-embedded or on-wafer NF versus packaged NF on the same die revision
- Measured input match and NF with ESD enabled vs a known ESD-less evaluation path (if the vendor provides one)
- Whether your matching network was tuned to the packaged S-parameters or to an ideal Gamma_opt from a core-only model

3. Habit
Never close a "we are 0.5 dB worse than the datasheet NF" ticket without asking whether the datasheet NF includes ESD and package. Archive ESD-included vs core-only notes next to every front-end NF claim. If the package S-parameters already contain the clamp, matching to a core-only Gamma_opt invents loss that looks like a process problem.

4. Practical gate
- Ask the vendor: NF and S-parameters with ESD on, package PN named
- On the board, treat the ESD network as part of the first matching section, not as "protection after RF"
- If you need both hard ESD and best NF, budget the clamp IL explicitly; do not hide it in "implementation loss"

Discussion: on your last NF miss versus datasheet, how much landed in ESD/package versus matching — and do you require ESD-included S-parameters before freezing a front-end budget?

— rfforum 编辑部 / Editorial Team
L Admin
Addendum — front-end NF traveler: write whether the S-parameters and NF include the ESD cell, and name the package PN, before you freeze a cascade budget. Matching to a core-only Gamma_opt against packaged parts invents 0.3–0.8 dB of “process” loss that was really the clamp.

Treat the ESD network as the first matching section, not as protection after RF. If you need both hard ESD and best NF, budget the clamp insertion loss explicitly; do not hide it in “implementation loss.”

Open: on your last NF miss versus datasheet, how much landed in ESD/package versus matching — and do you require ESD-included S-parameters before freezing the front-end budget?

— rfforum 编辑部 / Editorial Team
N Member
We've tested a similar approach - decent stability, but BOM cost goes up a bit.
R Member
On my last NF miss against the datasheet, the split was roughly two-thirds package and ESD, one-third matching — and I only found that out by measuring the same die in two forms, which is the part I now refuse to skip.

The datasheet NF was measured on-wafer, with the ESD cell de-embedded from the reported number, while my board used the packaged part with the full ESD network in front of the first stage. On the bench, the packaged part showed about 0.5 dB worse NF than the on-wafer number at the same bias, and the input match had also shifted by a few tenths of a dB in return loss. When I re-tuned the matching network to the packaged S-parameters, with the ESD cell included in the model, I recovered about 0.2 dB. The remaining 0.3 dB was the clamp itself: its shunt capacitance and series loss were simply in the signal path, and no amount of matching could remove loss that sits before the first gain stage.

That experience changed how I read a datasheet. I no longer accept an NF number without asking three things: does it include the ESD cell, which package was used, and was it measured on-wafer or packaged on the same die revision. If the vendor cannot answer those, I budget an extra 0.3 to 0.5 dB in the cascade for the ESD and package, and I write that number into the link budget explicitly instead of hiding it in implementation margin. I also ask for packaged S-parameters with the ESD on, and I tune the first matching section to those, not to a core-only Gamma_opt, because matching to the wrong reference impedance invents loss that looks like a process problem. The freeze list in this thread is exactly right — write down whether the S-parameters and NF include ESD, name the package, and keep the ESD-included and core-only notes next to every NF claim, so the next person does not have to rediscover the same 0.5 dB.
A Member
The traveler in the OP — whether NF and S-parameters include the ESD cell, and which package PN — is the right first question, and the on-wafer versus packaged split already discussed is exactly why a “process” miss of a few tenths of a dB can be a reference-plane miss. I would not treat any of those tenths of a dB already quoted as a universal budget. Two items still sit outside that freeze list and can move NF after the die and package are already understood.

The first is board-level protection that never appears in the vendor’s packaged S-parameters. A TVS, polymer suppressor, or spark gap at the connector, plus the via and ground return of that clamp, is another shunt C and a lossy path in front of the first stage. It shifts Gamma_opt again and adds insertion loss that no amount of retuning to the IC’s packaged S-parameters will remove. If the board also has an antenna bias-T or a leakage path that lets DC sit on the pin, the on-die diodes’ junction capacitance becomes bias-dependent: match and NF then walk with supply sequencing, antenna static drain, or a strong blocker that self-biases the clamp. That looks like a “hot NF” or blocker problem in the LNA core when it is still the clamp.

The second is the ESD return inductance and the fixture used to take the packaged S-parameters. A short, wide RF ground from the clamp to the reference plane is part of the matching section; a long return or a shared digital ground via adds series L that the vendor’s evaluation board may not have. When comparing on-wafer, packaged, and your board, name the de-embed method and whether the board-level clamp was in the path. Budget the discrete protection explicitly, the same way the thread already asks to budget on-die clamp insertion loss, and do not fold it into implementation loss.

If NF is still high after matching to ESD-included packaged S-parameters, look at the board clamp, pin DC, and ground return before opening a process ticket.
L Admin
@RF_Bench_Notes and @ARM_Linux_Lee — soft thank. The 0.5 dB datasheet miss is now split on the traveler, and the board-level clamp sits outside the vendor’s packaged S-parameters.

Locked keepers from RF_Bench_Notes:
1. Split on the last miss: roughly two-thirds package and ESD, one-third matching, found only by measuring the same die in two forms. Datasheet NF was on-wafer with the ESD cell de-embedded; the board used the packaged part with the full ESD network in front of the first stage. Packaged NF was about 0.5 dB worse at the same bias, and the input match had shifted by a few tenths of a dB. Re-tuning to packaged S-parameters with the ESD cell included recovered about 0.2 dB; the remaining 0.3 dB was the clamp itself — shunt C and series loss in front of the first gain stage, which no amount of matching can remove.
2. Three questions before accepting an NF number: does it include the ESD cell, which package was used, and was it measured on-wafer or packaged on the same die revision. If the vendor cannot answer, budget an extra 0.3–0.5 dB in the cascade explicitly, and tune the first matching section to packaged, ESD-included S-parameters, not to a core-only Gamma_opt.

Locked keepers from ARM_Linux_Lee:
3. Board-level protection (TVS, polymer suppressor, spark gap at the connector, plus its via and ground return) never appears in the vendor’s packaged S-parameters. It is another shunt C and a lossy path in front of the first stage. A bias-T or a leakage path that lets DC sit on the pin makes the on-die diodes’ capacitance bias-dependent — match and NF then walk with supply sequencing, antenna static drain, or a strong blocker that self-biases the clamp, and it looks like a “hot NF” or core-blocker problem when it is still the clamp.
4. ESD return inductance and the fixture used to take the packaged S-parameters belong on the same comparison. A long return or a shared digital ground via adds series L the evaluation board may not have. Name the de-embed method and whether the board-level clamp was in the path. Budget discrete protection the same way this thread already budgets on-die clamp insertion loss; do not fold it into implementation loss.

If NF is still high after matching to ESD-included packaged S-parameters, look at the board clamp, pin DC, and ground return before opening a process ticket.

— rfforum 编辑部 / Editorial Team
广告 AdRenesas 瑞萨微控制器芯片产品照片Renesas Electronics微控制器、模拟与射频/时钟方案 · MCUs, analog, RF and timing访问官网 →免费展示 · 资料来自公开信息 · 如需撤下请联系 ads@rfforum.com
广告 Ad广告位招商 / Advertise hereads@rfforum.com · 刊例 Rate cardT2 · 728×90
Log in or register to reply.