Applications
Anything that listens, times or converts is limited by the noise on its supply. Each section covers the problem, what Harmony changes, where to put it, and the measurements behind it.
AI data center, optics and compute
Supply noise turns into jitter on every clock and data edge. Quiet rails mean timing margin for TIAs, laser drivers, SerDes, DDR and HBM.
| The problem | Next-generation links run 200G per lane over active copper cable and 400G per lane over optics. Their TIAs, laser drivers, equalizers, cable drivers, SerDes and clocks all share supply rails. As process nodes shrink, operating voltages fall, so a fixed amount of supply noise takes a bigger share of a smaller margin. Through power-supply-induced jitter it becomes timing error and bit errors. |
|---|---|
| What Harmony changes | Harmony shunts the rail’s noise to ground at the point of load. A market leader validated it in a high-speed laser source: a 0.95 V, 4 A switching rail measured flat on a 59 GHz Keysight UXR oscilloscope. With 200 mVp-p injected on its supply, a differential XO’s phase noise improved by 18 to 30 dB from 20 kHz to 2 MHz. |
| Where to place it | Type-B at the point of load for TIA, laser-driver, SerDes, PLL and memory-interface rails. Use dual Type-B inside the package or module when you need suppression to 20 GHz. |
| Typical circuits | Timing, TIA, laser source and driver, equalizer, cable driver, SerDes, photodiode, ACC, AEC, AOC, OSFP, QSFP, GPU, TPU, DDR, HBM, memory controller, SSD |
| Evidence | Case study: a 625 MHz clock at 15.73 fs Case study: a flat rail in a high-speed laser source Case study: clock phase noise with 200 mVp-p injected Harmony for 2.4T and 3.2T copper and optical pluggables Harmony for memory: PCIe Gen5/6 SSDs and DDR5/6 Why bypass networks top out |
Wireless, 5G and RF front ends
Supply noise becomes phase noise and spurs. Filter the oscillator and front-end rails, and a commodity supply behaves like a lab bias.
| The problem | An oscillator’s frequency shifts slightly with its supply, so ripple becomes spurs and broadband noise raises the phase-noise floor. In direct-sampling receivers that jitter caps ADC SNR. Front-end modules surrounded by LC filters pick up high-Q resonances instead. |
|---|---|
| What Harmony changes | A 122.88 MHz VCXO on a noisy 3.3 V rail went from 233 fs to 20 fs RMS jitter, against 18 fs in the lab. Two Skyworks RF front-end modules dropped from 16 decoupling parts to 3 and operate at the noise floor. NYU WIRELESS measured 12.5 dB less supply noise at a 2.4 GHz receiver. |
| Where to place it | On VCXO, TCXO, OCXO and PLL supplies, on the FEM VCC rails, and between the PA or digital rails and the LO chain. |
| Typical circuits | Timing, digitizer, wireless backhaul, SoC, LNA, 5G, 6G |
| Evidence | Case study: 5G clock jitter Case study: two Skyworks FEMs, 16 parts to 3 Case study: NYU WIRELESS BLE link |
GNSS and positioning
Satellite signals arrive below the noise floor. Clean supplies cut worst-case error tenfold and keep tracks on the trail.
| The problem | GPS L1 signals arrive about 17 dB below thermal noise in a 2 MHz bandwidth. Under trees or buildings, supply noise in the LNA, synthesizer and TCXO competes directly with the satellites, causing position scatter, dropouts and slow fixes. |
|---|---|
| What Harmony changes | In a 24-hour side-by-side test, worst-case error fell from 2.5 km to about 200 m. On a 5 km forest trail, two watches with the same GNSS chipset recorded 5.27 km without TransSiP PI and 5.03 km with it. TransSiP PI also runs in a dual-band L1/L5 module built on Sony’s CXD5610. |
| Where to place it | On the GNSS module or LNA, TCXO and synthesizer supplies. |
| Typical circuits | GPS and GNSS receivers, sports watches, bike computers, asset and pet trackers |
| Evidence | Case study: navigation that stays on the trail 10× smaller GNSS position errors |
Precision analog and ultra-low-noise LDOs
A further 10 to 20 dB below an LDO’s own noise floor, including its 1/f region, goes straight into SNR for ADCs, DACs, audio and sensors.
| The problem | Even the best LDOs have a 1/f noise region and a PSRR that rolls off with frequency. Their noise floor becomes the converter’s or the sensor’s noise floor. |
|---|---|
| What Harmony changes | With Nokia Bell Labs, two ADI LT3046 LDOs in series went from spurs at −100 dBm to a clean −130 dBm floor from 1 Hz to 6 GHz. An ADM7150 in a defense radio showed 20 dB less 1/f noise, reaching power-off levels. A TI TPS7A94 reached the analyzer floor from 3 Hz to 6 GHz in measurements with Nokia Bell Labs. In the signal chain, Harmony’s native 700 Hz cutoff suits slow sensors such as geophones, without power-hungry digital filtering. |
| Where to place it | At the LDO’s VIN, VOUT and reference node, at ADC, DAC and AFE supply pins, or in the signal path of sub-kilohertz sensors. |
| Typical circuits | ADC, DAC, FPGA, audio, LEDs, photodetectors, geophones, MEMS |
| Evidence | Case study: two LT3046s, every spur removed Case study: TPS7A94 at the analyzer floor Case study: ADM7150 at power-off levels Harmony with ultra-low-noise LDOs |
Power conversion and PDN design
Keep the switcher’s efficiency and drop the LDO post-regulator. Switcher ripple falls from millivolts to microvolts.
| The problem | Switch-mode supplies are efficient and noisy. Linear regulators are quieter, but they waste headroom. Classic target-impedance PDNs need a ladder of capacitors, they resonate, and each design is tuned to its own load. |
|---|---|
| What Harmony changes | At the output of an ADI LT8622S, ripple fell from 10 mVp-p to 4.48 µVp-p. A noisy 3.3 V rail reached the analyzer floor with three parts, where TI’s 15-part 6 GHz filter left it “still noisy”. A PDN resonance of 1.13 Ω fell to 153 µΩ. |
| Where to place it | Pin 1 close to the power inductor and the feedback node for SNJ conditioning, or Type-B at the point of load when the rail picks up radiated energy. |
| Typical circuits | PMIC, SMPS, LDO, VRM, energy harvesting, USB power |
| Evidence | Case study: LT8622S, 2,232× lower ripple Case study: three parts against a 15-part filter Why the classic target-impedance approach falls short |
Biomedical and optical sensing
A wrist-worn heart-rate monitor with one LED and one detector matched a chest strap once its bias was clean.
| The problem | Arterial pulses change the optical signal by about 0.086 dB or less. That is easily buried by supply noise on the LEDs, photodetector, AFE and current ADC. The usual fix of more LEDs and more current costs power and still falls short. |
|---|---|
| What Harmony changes | TransSiP PI-enhanced optical monitors reached chest-strap accuracy with a single LED and detector. In Japan, MEDIROM’s body-heat-powered MOTHER bracelet uses four TransSiP PI products, and PI-enabled health bracelets are used in hospitals and clinics. |
| Where to place it | On the LED driver, photodetector, AFE and current-ADC bias rails. |
| Typical circuits | Optical heart-rate monitors, wearables, biosensors |
| Evidence | Case study: a body-heat-powered health tracker |
Space, defense and quantum
Mission-critical radios, satellites and cryogenic electronics need clean rails and very few parts.
| The problem | Space, defense and quantum systems combine tight size, weight and power budgets with zero tolerance for noise-induced errors. Every decoupling part is another solder joint to qualify. |
|---|---|
| What Harmony changes | A defense radio’s ADM7150 LDO reached power-off noise levels. A mission-critical RF board went from 16 decoupling parts to 3. We are collaborating with industry experts on DC-to-20 GHz supply-noise elimination for quantum computing at 4 K. |
| Where to place it | Type-B on every sensitive rail, and dual Type-B where suppression must reach 20 GHz. |
| Typical circuits | Satellites, modular spacecraft, defense radios, quantum computing |
| Evidence | Case study: supply-noise elimination at 4 K Case study: ADM7150 at power-off levels |
Also used in
Energy harvesting, System-in-Package, Consumer electronics, Hi-fi audio, Remote sensing. Tell us about your system and an applications engineer will suggest a placement.