Fan noise rises with the fifth power of speed. How silent cooling with solid-state plasma actuators meets medical, audio, lab and in-cabin noise specs.
A 40 mm axial fan with seven blades turning at 6,000 rpm emits a discrete tone at 700 Hz. The arithmetic is not subtle: blade pass frequency is the number of blades multiplied by the rotational speed in rpm and divided by 60 [1]. That tone, plus its harmonics, sits in a band where A-weighting barely discounts anything, and it is the reason a laptop, an ultrasound cart or a rack-mounted audio interface announces itself the moment its thermal controller decides the die is running warm.
Most cooling discussions treat noise as a comfort issue that gets traded away when temperatures climb. In a growing set of products it is not a comfort issue. It is a line in the specification, verified on a test bench to ISO 7779 [2], and a design that misses it does not ship.
Silent cooling means moving heat away from electronics without a rotating or oscillating mechanical element, so the thermal solution contributes no blade passing tone, no bearing noise and no structure-borne vibration to the product's acoustic signature. A dielectric barrier discharge plasma actuator does this by ionising a shallow layer of air and dragging neutral air along a surface, producing a wall jet with nothing that spins.
"Fan noise" is five different mechanisms sharing one enclosure, and they respond to different fixes.
| Source | Character | Set by | Usual mitigation |
|---|---|---|---|
| Blade passing | Tonal, at BPF and its harmonics | Blade count times rpm [1] | Uneven blade spacing, larger blade-to-obstruction gap |
| Broadband turbulence | Hiss, wide spectrum | Local flow velocity, inlet obstruction | Lower speed, cleaner intake path |
| Motor commutation | Tonal, at drive and pole harmonics | Commutation scheme, pole count | Sinusoidal rather than trapezoidal drive |
| Bearing | Broadband, degrading with age and orientation | Load, lubricant, wear state | Fluid dynamic bearings, vertical shaft |
| Structure-borne | Tonal, at rotor imbalance frequency | Imbalance, mount stiffness | Elastomeric isolators, stiffer chassis |
The last row is the one that escapes acoustic budgets most often. A fan that measures acceptably in free field can excite a chassis panel into radiating far more sound than the rotor itself. This path is important enough to have its own measurement standard: ISO 10302-2 specifies how to determine the vibration a small air-moving device induces in the structures used in information technology and telecommunications equipment [3]. Chassis resonance is a property of the assembly, not of the fan, which is why it usually surfaces late and expensively.
Two sounds at the same A-weighted level are not equally objectionable. The cochlea performs a spectral decomposition, so a concentrated tonal component produces a localised peak in the excitation pattern that stands out against the broadband background even when it carries a small fraction of the total acoustic energy [4].
Standards bodies encode this. ECMA-74 specifies a method for determining whether an equipment's noise emission contains prominent discrete tones [5]. Environmental and product acoustics regimes then apply a penalty to tonal sources. Listening experiments reported at ICBEN 2023 found tonal penalties ranging from zero up to 12 dB depending on tonal audibility and frequency, with the largest values, 8 to 12 dB, at the highest audibility tested, and no penalty at all for low-frequency tones in the 50 to 110 Hz band [6]. The same work cautions that fixed penalty models are not justified, which is itself useful: the cost of a tone depends on where it sits and how far it protrudes, not on a constant you can look up.
The practical consequence for anyone doing fan noise reduction in electronics is that broadband level is the easy half of the problem. A product can pass its dBA target and still be rejected in listening review because a 700 Hz whine is audible across a room.
The reason thermal engineers reach for fan speed and acoustic engineers dread it is a scaling law. All else equal, fan noise goes as 50 log(rpm) [7]. Fifty log is the tenth-of-a-bel expression of the fifth power of speed, while volumetric airflow rises only in proportion to speed.
| Speed increase | Extra airflow | Extra sound power |
|---|---|---|
| +10% | +10% | about 2.1 dB |
| +25% | +25% | about 4.8 dB |
| +50% | +50% | about 8.8 dB |
| +100% | +100% | about 15.1 dB |
Doubling airflow to chase a hot spot costs roughly 15 dB. Acoustics collapses long before airflow does. This is why so many quiet products end up thermally throttled instead: the fan curve has headroom and the noise budget does not.
It also explains why the hot spot matters more than the enclosure average. A single optical module, power stage or memory package sitting a few kelvin over its limit can force the whole chassis fan up a step, paying the full acoustic penalty for a local problem. Targeting the boundary layer over that one component, rather than raising bulk flow, is the argument for ionic wind as a solid-state cooling approach.
A plasma actuator has no rotor, no bearing and no diaphragm. There is no blade pass frequency because there are no blades, no commutation tone because there is no motor, no bearing spectrum because there is no bearing, and no rotor imbalance to inject into the chassis. Published work on ionic wind for electronics states plainly that the process "requires no moving parts and produces no acoustic noise" [8]. Peak velocities of 1.5 to 2 m/s are reported in the impingement zone below the emitter in that study [8], which is the right order for boundary-layer work and the wrong order for ventilating a large volume.
This also distinguishes the approach from the other mechanical-free options. A piezoelectric cooler removes the bearing but keeps an oscillating membrane, so it keeps a driven tone at its resonance; the comparison with piezo fans turns on exactly that point. The trade against fans on flow and pressure is covered in the fans versus ionic wind comparison.
| Environment | Governing criterion | Consequence for cooling |
|---|---|---|
| Patient rooms and wards | WHO guideline of 30 dB LAeq and 40 dB LAmax indoors for sleep disturbance [9] | Bedside equipment must be quieter than the room target |
| Imaging and treatment rooms | WHO gives no number, only "as low as possible" [9] | Acoustics judged by clinician acceptance |
| Recording and broadcast | NC 15 to 20 typical target [10] | Any equipment tone is captured on the microphone |
| Quiet office, home devices | NC 25 to 30 typical target [10] | Tonal content dominates complaints |
| Hospitals, hotels, classrooms | NC 30 to 35 typical target [10] | Broadband is tolerable, tones are not |
| Laboratory and metrology | Vibration limit, not just sound | Structure-borne path is the binding constraint |
| In-cabin automotive | Interior sound quality target per programme | Electronics must vanish under road noise at idle |
Metrology instruments are the sharpest case. An interferometer, a probe station or a mass comparator cares less about audible level than about the vibration a fan injects into its own frame, which is measured in the ISO 10302-2 sense [3] rather than in dBA. Removing the rotating mass removes the excitation, not just the sound.
A plasma actuator is not silent in an absolute sense, and it is dishonest to claim otherwise. Two things can be heard.
The discharge itself is a repetitive electrical event. Driven by a kilovolt-class AC supply, it produces a small pressure disturbance each half cycle, so acoustic content appears at the drive frequency and its harmonics. Push the drive fundamental well above the audible band, which the physics permits since audibility runs from roughly 20 Hz to 20 kHz [11], and what remains is a faint hiss at close range rather than a room-filling tone.
The drive electronics are the more likely offender. Switch-mode converters produce audible noise through magnetostriction in inductor and transformer cores, and through electrostrictive and piezoelectric forces in ceramic capacitors, the effect commonly called coil whine [11]. If the switching fundamental or its low-order harmonics land inside the audible band, a nominally silent cooler acquires a tone from its power stage. This is a solvable design problem, governed by switching frequency selection, magnetics construction, potting and layout, but it has to be designed in rather than assumed away.
Beyond acoustics, the honest limits stand. Ionic wind acts as a surface convection enhancer rather than a bulk heat extractor [8], with electrical-to-fluid conversion efficiency around 1 to 2 per cent [12]. If the requirement is high volumetric flow through a deep heat sink, a fan is still the correct component and the answer is a quieter fan, not no fan. There is also ozone generation in air, waveform and geometry dependent, discussed in the ionic wind reliability and ozone article. The strongest case is hybrid: a slow, quiet fan for bulk ventilation plus local actuation on the components that would otherwise set the fan speed, which is the configuration behind fanless and low-noise GPU cooling work and the Y-Flow thermal product line. If you have an acoustic specification you are currently missing, tell us the number and the hot spot.
Is silent cooling with a plasma actuator completely noiseless? No. The discharge produces acoustic content at the drive frequency and its harmonics, and the high-voltage drive electronics can produce coil whine if the switching frequency falls in the audible band. What it does remove is the blade passing tone, motor commutation tone, bearing noise and rotor-imbalance vibration that dominate fan acoustics.
Why does speeding up a fan make it so much louder? Because fan noise scales as 50 log(rpm), which is the fifth power of speed, while airflow rises only in proportion to speed. Doubling airflow costs roughly 15 dB, so the acoustic budget runs out well before the aerodynamic one.
Why is a tone more annoying than hiss at the same dBA? The ear resolves a tone as a localised peak in the cochlear excitation pattern, so it stands out against broadband background even at low relative energy. Measured penalties for tonal noise have ranged up to 12 dB depending on audibility and frequency.
Which markets treat quiet as a hard specification? Medical imaging and patient rooms, recording and broadcast, laboratory and metrology instruments, in-cabin automotive electronics, premium consumer devices and quiet-office equipment. These are governed by WHO indoor guidelines, NC or RC room criteria, or programme-level sound quality targets rather than by preference.
Can a plasma actuator replace the fan entirely? Rarely. Ionic wind produces metres per second, not tens of metres per second, and acts on the boundary layer rather than ventilating a volume. The realistic quiet cooling solution is a slower fan plus local actuation on the hot spots that were setting the fan speed.