Harmony for memory: clean power for PCIe Gen5/6 SSDs and DDR5/6
Merged IO rails, up to 16 ONFI channels and shrinking voltage margins make power integrity a speed limit for SSDs and DDR. Where the noise comes from, and what a 5.7 mm² filter changes.
Measurements, design notes and the physics behind them, from our applications engineers.
Merged IO rails, up to 16 ONFI channels and shrinking voltage margins make power integrity a speed limit for SSDs and DDR. Where the noise comes from, and what a 5.7 mm² filter changes.
2.4T and 3.2T pluggables run 300G and 400G per lane, leaving less timing margin for supply noise than ever. Where Harmony goes in ACC, AEC, AOC, OSFP, ELSFP, LPO and NPO.
The best LDOs still have a 1/f region, a PSRR that rolls off, and a real-world noise floor above their datasheet. The newest result, from Nokia Bell Labs, takes two ADI LT3046s to a −130 dBm floor.
We put a 15-part, 6 GHz broadband filter and a 3-part Harmony Type-B on the same noisy 3.3 V rail. One was just quiet enough to pass EMI. The other reached the analyzer’s noise floor.
We measured an ADI LT8622S on two oscilloscopes and then added one Harmony PI Filter. The ripple a 1.5 GHz scope misses is the ripple that matters.
Two identical GPS receivers shared one antenna and one noisy USB supply for 24 hours. The only difference between them was the supply bias, and it changed the worst-case error from 2.5 km to about 200 m.
Capacitors resonate, ferrite beads saturate and LDO rejection rolls off. This is the physics behind the 80–90 dB ceiling of conventional PDN filtering, and what happens when one part removes it.
Working with oscillator maker Rakon, TransSiP improved a 5G reference clock about six-fold while it ran from a noisy supply. Here is why that is worth about 15 dB of SNR in a direct-sampling receiver.