Reservoir computing was achieved by constructing a network of virtual nodes multiplexed in time and sharing a single silicon beam exhibiting a classical Duffing non-linearity as the source of nonlinearity. The delay-coupled electromechanical system performed well on time series classification tasks, with error rates below 0.1% for the 1st, 2nd, and 3rd order parity benchmarks and an accuracy of $(78\pm 2)\text{\%}$ for the TI-46 spoken word recognition benchmark. As a first demonstration of reservoir computing using a non-linear mass-spring system in MEMS, this result paves the way to the creation of a new class of compact devices combining the functions of sensing and computing.

As it is getting increasingly difficult to achieve gains in the density and power efficiency of microelectronic computing devices because of lithographic techniques reaching fundamental physical limits, new approaches are required to maximize the benefits of distributed sensors, micro-robots or smart materials. Biologically-inspired devices, such as artificial neural networks, can process information with a high level of parallelism to efficiently solve difficult problems, even when implemented using conventional microelectronic technologies. We describe a mechanical device, which operates in a manner similar to artificial neural networks, to solve efficiently two difficult benchmark problems (computing the parity of a bit stream, and classifying spoken words). The device consists in a network of masses coupled by linear springs and attached to a substrate by non-linear springs, thus forming a network of anharmonic oscillators. As the masses can directly couple to forces applied on the device, this approach combines sensing and computing functions in a single power-efficient device with compact dimensions.

This study presents the design, fabrication, and test of a micro accelerometer with intrinsic processing capabilities, that integrates the functions of sensing and computing in the same MEMS. The device consists of an inertial mass electrostatically coupled to an oscillating beam through a gap of 8 μm. The motion of the inertial mass modulates an AC electrostatic field that drives the beam in its non-linear regime. This non-linearity is used to implement machine learning in the mechanical domain, using reservoir computing with delayed feedback to process the acceleration information provided by the inertial mass. The device is microfabricated on a silicon-on-insulator substrate using conventional MEMS processes. Dynamic characterization showed good accelerometer functionalities, with an inertial mass sensitivity on the order of 100 mV/g from 250 to 1300 Hz and a natural frequency of 1.7 kHz. In order to test the device computing capabilities, two different machine learning benchmarks were implemented, with the inputs fed to the device as accelerations. The neuromorphic MEMS accelerometer was able to accurately emulate non-linear autoregressive moving average models and compute the parity of random bit streams. These results were obtained in a test system with a non-trivial transfer function, showing a robustness that is well-suited to anticipated applications. [2019-0238]

In-sensor computing could become a fundamentally new approach to the deployment of machine learning in small devices that must operate securely with limited energy resources, such as wearable medical devices and devices for the Internet of Things. Progress in this field has been slowed by the difficulty to find appropriate computing devices that operate using physical degrees of freedom that can be coupled directly to degrees of freedom that perform sensing. Here we leverage reservoir computing as a natural framework to do machine learning with the degrees of freedom of a physical system, to show that a micro-electromechanical system can implement computing and the sensing of accelerations by coupling the displacement of suspended microstructures. We present a complete wearable system that can be attached to the foot to identify the gait patterns of human subjects in real-time. The computing efficiency and the power consumption of this in-sensor computing system is then compared to a conventional system with a separate sensor and digital computer. For similar computing capabilities, a much better power efficiency can be expected for the highly-integrated in-sensor computing devices, thus providing a path for the ubiquitous deployment of machine learning in edge computing devices.

This paper investigated the effect of heat spreading on the boiling of the Novec 649™ for two-phase immersion cooling of electronics. Reference pool boiling tests were performed by attaching a 25.4 mm by 25.4 mm square copper plate to a same-sized heater, thus minimizing lateral heat spreading. Experimental measurements showed that the critical heat flux (CHF) happened at a heat flux of 17.4±0.8 W/cm2. Then, lateral heat spreading through the heat spreader was studied by attaching larger (47 mm by 47 mm) spreaders with four different thicknesses to the copper plate. With an increase in the integrated heat spreader (IHS) thickness from 1 mm to 6 mm, the CHF increased by more than 60% at the saturation condition. One plate was a 1 mm-thick IHS removed from a commercial microprocessor. In this case, the CHF happens at 8.6 W/cm2 (50% lower compared to the reference case) in the saturation condition. At CHF, the boiling can be observed on the whole surface, with columns and slugs regime at the center and the fully developed nucleate boiling regime at the edges. This nonuniform boiling was more pronounced in subcooled conditions, in which the CHF occurred at the center while there were regions at the edges that had no boiling. Finally, the performance of a microporous-coated IHS (with 3.15 mm thickness) was compared to the 6 mm thick IHS. The thermal resistance was almost equal for powers above 200 W. This indicates that lateral heat spreading is a critical parameter for the thermal design of immersion cooling along with microporous coating.

In a two-phase immersion cooling system, boiling on the spreader surface has been experimentally found to be nonuniform, and it is highly related to the surface temperature and the heat transfer coefficient. An experimentally obtained temperature-dependent boiling heat transfer coefficient has been applied to a numerical model to investigate the spreader's cooling performance. It is found that the surface temperature distribution becomes less uniform with higher input power. But it is more uniform when the thickness is increased. By defining the characteristic temperatures that represent different boiling regimes on the surface, the fraction of the surface area that has reached the critical heat flux has been numerically calculated, showing that increasing the thickness from 1 mm to 6 mm decreases the critical heat flux reached area by 23% at saturation liquid temperatures. Therefore, on the thicker spreader, more of the surface is utilized for nucleate boiling while localized hot regions that lead to surface dry-out are avoided. At a base temperature of 90 °C, the optimal thickness is found to be 4 mm, beyond which no significant improvement in heat removal can be obtained. Lower coolant temperatures can further increase the heat removal; it is reduced from an 18% improvement in the input power for the 1 mm case to only 3% in the 6 mm case for a coolant temperature drop of 24 °C. Therefore, a tradeoff exists between the cost of maintaining the low liquid temperature and the increased heat removal capacity.