Thermal noise is a fundamental limitation in the performance of MEMS and NEMS resonators and bolometers, particularly in applications requiring high sensitivity, such as IoT and biomedical sensors. Minimizing this noise involves a combination of material selection, environmental control, and active stabilization techniques. The following sections detail strategies for thermal noise suppression, along with their associated trade-offs.

Low-stress materials like silicon nitride (SiNx) are widely used in MEMS/NEMS resonators due to their excellent mechanical properties and low intrinsic dissipation. The stress state of the material directly influences the quality factor (Q) of the resonator, which is critical for reducing thermomechanical noise. High-tensile-stress SiNx films exhibit Q factors exceeding 1 million in vacuum, significantly lowering the Brownian motion-induced noise floor. The reduction in mechanical loss is attributed to the suppression of phonon-phonon interactions and defect-related dissipation. However, achieving ultra-low-stress uniformity across large wafers remains a challenge, as residual stress gradients can introduce frequency instability and mode hybridization. Trade-offs include the difficulty in depositing stress-controlled films at low temperatures, which may limit integration with CMOS processes.

Vacuum packaging is essential for minimizing gas damping, a dominant noise source in ambient conditions. For MEMS resonators operating at atmospheric pressure, viscous damping drastically reduces Q factors to the order of hundreds or thousands. By maintaining pressures below 1 mTorr, viscous damping is effectively eliminated, leaving thermoelastic damping and anchor losses as the primary dissipation mechanisms. The required vacuum level depends on the device dimensions and operational frequency; smaller gaps and higher frequencies demand lower pressures to avoid squeeze-film damping. Hermetic sealing techniques, such as wafer-level bonding with getter materials, ensure long-term vacuum stability. However, vacuum packaging increases fabrication complexity and cost, while also introducing challenges in electrical feedthrough design. Additionally, the thermal conductance of the device is reduced in vacuum, complicating heat dissipation in bolometers.

Active feedback cooling is a powerful method to suppress thermal noise beyond the limits imposed by passive techniques. By employing a feedback loop that detects resonator motion and applies a counteracting force, the effective temperature of the mechanical mode can be reduced. Optical or electrical transduction schemes are typically used for displacement detection, while electrostatic or piezoelectric actuation provides the cooling force. Experimental implementations have demonstrated mode cooling from room temperature to below 10 K in SiNx resonators. The cooling efficiency depends on the loop delay and feedback gain, with higher gains providing greater noise suppression but risking instability. A key trade-off is the increased power consumption of the feedback electronics, which may be prohibitive for energy-constrained applications like IoT sensors. Furthermore, active cooling introduces additional electronic noise, which must be carefully managed to avoid degrading the overall signal-to-noise ratio.

In bolometers, thermal noise suppression is achieved through a combination of low-thermal-conductance supports and sensitive thermometric materials. Suspended structures with narrow beams minimize conductive heat loss, enhancing temperature sensitivity. Materials such as vanadium oxide (VOx) and amorphous silicon (a-Si) are favored for their high temperature coefficient of resistance (TCR), enabling sub-mK thermal resolution. However, reducing thermal conductance also increases the thermal time constant, limiting the sensor bandwidth. For instance, a bolometer with a thermal conductance of 10 nW/K may achieve a noise-equivalent power (NEP) of 1 pW/Hz½ but with a response time on the order of milliseconds. This trade-off necessitates careful optimization based on the application requirements—biomedical sensors may prioritize sensitivity over speed, while high-speed imaging demands faster response at the cost of higher NEP.

The choice of readout circuitry further influences noise performance. For capacitive MEMS resonators, low-noise transimpedance amplifiers with input-referred noise below 1 fA/Hz½ are critical for resolving small displacements. In resistive bolometers, bridge circuits or low-noise analog-to-digital converters (ADCs) must be optimized to avoid introducing additional electronic noise. Power consumption scales with bandwidth and resolution; a high-resolution ADC operating at 1 kS/s may consume microwatts, whereas MHz sampling rates can demand milliwatts. IoT applications often strike a balance by employing duty-cycling or event-driven sampling to conserve energy.

Integration with CMOS processes presents additional challenges and opportunities. Monolithic integration reduces parasitic capacitances and enables co-design of MEMS/NEMS structures with low-noise electronics. However, thermal budgets for post-CMOS MEMS processing must be carefully managed to avoid degrading transistor performance. Heterogeneous integration, such as bonding pre-fabricated MEMS chips to CMOS wafers, offers an alternative but introduces interconnect parasitics that can degrade signal integrity.

Emerging materials like 2D semiconductors and ultra-low-loss dielectrics may further push the limits of thermal noise suppression. Graphene membranes, for example, exhibit exceptionally high Q factors due to their atomic thinness and stiffness. However, challenges in reproducible fabrication and integration with conventional MEMS processes remain significant barriers to widespread adoption. Similarly, topological insulator materials show promise for reducing dissipation in nanomechanical systems but require further development to achieve practical device integration.

In summary, suppressing thermal noise in MEMS/NEMS resonators and bolometers requires a multi-faceted approach involving material engineering, environmental control, and advanced signal processing. Each technique presents inherent trade-offs between sensitivity, bandwidth, and power consumption, necessitating application-specific optimization. For IoT sensors, low-power operation may take precedence, while biomedical applications might favor ultimate sensitivity at the expense of higher energy use. Continued advancements in materials science and microfabrication techniques will further enhance the performance and applicability of these devices in noise-critical environments.