Nanomechanical network for computation

Abstract

A nano-electromechanical system (NEMS) oscillator network and a method of operating the same are disclosed. The NEMS oscillator network includes one or more network inputs configured to receive one or more input signals. The NEMS oscillator network also includes a plurality of NEMS oscillators coupled to the one or more network inputs. Each of the plurality of NEMS oscillators includes a NEMS resonator and generates a radio frequency (RF) output signal oscillating at a particular frequency and a particular phase. The NEMS oscillator network also includes a plurality of connections interconnecting the plurality of NEMS oscillators. The NEMS oscillator network also includes one or more network outputs coupled to the plurality of NEMS oscillators and configured to output one or more output signals.

Description

[0001] Cross-references to related applications

[0002] This application claims priority to U.S. Provisional Patent Application No. 62 / 841112, filed April 30, 2019, entitled “NANOMECHANICAL NETWORKS FOR COMPUTATION”, the disclosure of which is incorporated herein by reference in its entirety. Background Technology

[0003] From advanced machine learning to image processing in mobile phone camera systems, CPU-centric computing is declining. For example, in many smartphones, a significant portion of processing is currently not done by the central processing unit (CPU), but by dedicated chips (including graphics processing units, image signal processors, modems, etc.). The same problem is prominent in machine learning, where combinatorial optimization and low-precision matrix multiplication require substantial computation.

[0004] For many of these applications, the communication latency between the processor and memory (known as the von Neumann bottleneck) is particularly severe due to the large amounts of data being processed and the significant memory usage. While complementary metal-oxide-semiconductor (CMOS) accelerators have achieved faster performance through parallelization, the uneven distribution of load among parallel cores and the synchronous nature of the current architecture inevitably lead to slower speeds. Given the potential end of CMOS scaling, there is a need to develop new and alternative large-scale, fast, and low-power solutions. Summary of the Invention

[0005] This invention generally relates to methods and systems for performing computations using nanoelectromechanical systems (NEMS) oscillator networks. More specifically, embodiments of the invention relate to the structure of such networks and methods for training them for specific tasks.

[0006] A system of one or more computers can be configured to perform specific operations or actions by installing software, firmware, hardware, or combinations thereof on the system (causing the system to perform these actions in operation). One or more computer programs can be configured to perform specific operations or actions by including instructions that, when executed by a data processing device, cause that device to perform these actions. A general aspect includes a nanoelectromechanical system (NEMS) oscillator network comprising: one or more network inputs configured to receive one or more input signals. The nanoelectromechanical system also includes a plurality of NEMS oscillators coupled to the one or more network inputs, wherein each of the plurality of NEMS oscillators generates a radio frequency (RF) output signal oscillating at a specific frequency and a specific phase. The nanoelectromechanical system also includes a plurality of connections interconnecting the plurality of NEMS oscillators, such that each of the plurality of NEMS oscillators is connected to at least one other NEMS oscillator among the plurality of NEMS oscillators via at least one of the plurality of connections. The nanoelectromechanical system also includes one or more network outputs coupled to the plurality of NEMS oscillators, wherein the one or more network outputs are configured to output one or more output signals. Other embodiments of this aspect include corresponding computer systems, apparatuses, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the method.

[0007] Implementations may include one or more of the following features: A NEMS oscillator network, wherein each of a plurality of NEMS oscillators includes a NEMS resonator. A NEMS oscillator network, wherein each of a plurality of NEMS oscillators includes a feedback path between the output and input of the NEMS resonator. A NEMS oscillator network, wherein each of a plurality of connections includes coupling weights of a plurality of coupling weights, the coupling weights causing an amplitude variation in the RF output signal through the connection. A NEMS oscillator network, wherein each of the plurality of coupling weights also causes a phase variation in the RF output signal through the connection. A NEMS oscillator network, wherein the RF output signal generated by each of the plurality of NEMS oscillators is transmitted through at least one of the plurality of connections. A NEMS oscillator network, wherein the NEMS oscillator network is trained by a training process, wherein the training process provides one or more input signals at one or more network inputs. The NEMS oscillator network may also include reading one or more output signals at one or more network outputs. The NEMS oscillator network may also include performing a comparison between one or more output signals and one or more reference output signals. The NEMS oscillator network may also include modifying the NEMS oscillator network based on the comparison. Implementations of the described technology may include hardware, methods or processes, or computer software on a computer-accessible medium.

[0008] One general aspect includes a method of operating a network of NEMS oscillators, the method comprising: receiving one or more input signals at one or more network inputs of the NEMS oscillator network. The method further includes generating a radio frequency (RF) output signal for each of a plurality of NEMS oscillators in the NEMS oscillator network, the NEMS oscillators being coupled to the one or more network inputs, and wherein the NEMS oscillator network includes a plurality of connections interconnecting the plurality of NEMS oscillators, such that each of the plurality of NEMS oscillators is connected to at least one other NEMS oscillator among the plurality of NEMS oscillators via at least one of the plurality of connections. The method further includes outputting one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs being coupled to the plurality of NEMS oscillators. Other embodiments of this aspect include corresponding computer systems, apparatuses, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the method.

[0009] Implementations may include one or more of the following features: A NEMS oscillator network, wherein each of a plurality of NEMS oscillators includes a NEMS resonator. In this method, each of the plurality of NEMS oscillators includes a feedback path between the output and input of the NEMS resonator. In this method, each of the plurality of connections includes a coupling weight of a plurality of coupling weights that cause an amplitude change in the RF output signal through the connection. In this method, each of the plurality of coupling weights also causes a phase change in the RF output signal through the connection. In this method, the RF output signal generated by each of the plurality of NEMS oscillators is transmitted through at least one of the plurality of connections. Implementations of the described techniques may include hardware, methods, or processes, or computer software on a computer-accessible medium.

[0010] One general aspect includes a method for training a NEMS oscillator network, the method comprising: providing one or more input signals to one or more network inputs of the NEMS oscillator network. The method further includes generating a radio frequency (RF) output signal for each of a plurality of NEMS oscillators of the NEMS oscillator network that oscillates at a specific frequency and a specific phase, wherein the plurality of NEMS oscillators are coupled to the one or more network inputs, and wherein the NEMS oscillator network includes a plurality of connections interconnecting the plurality of NEMS oscillators, such that each of the plurality of NEMS oscillators is connected to at least one other NEMS oscillator among the plurality of NEMS oscillators via at least one of the plurality of connections. The method further includes reading one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs being coupled to the plurality of NEMS oscillators. The method further includes comparing the one or more output signals with one or more reference output signals. The method further includes modifying the NEMS oscillator network based on the comparison. Other embodiments of this aspect include corresponding computer systems, apparatuses, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the method.

[0011] Implementations may include one or more of the following features. In this method, each of a plurality of NEMS oscillators includes a NEMS resonator. In this method, each of the plurality of NEMS oscillators includes a feedback path between the output and input of the NEMS resonator. In this method, each of the plurality of connections includes a coupling weight of a plurality of coupling weights that causes an amplitude change in the RF output signal through that connection. A method of modifying the NEMS oscillator network based on comparison includes modifying the amplitude change of at least one of the plurality of coupling weights. In this method, each of the plurality of coupling weights also causes a phase change in the RF output signal through the connection. In this method, the RF output signal generated by each of the plurality of NEMS oscillators is transmitted through at least one of the plurality of connections. Implementations of the described techniques may include hardware, methods, or processes, or computer software on a computer-accessible medium.

[0012] Compared to conventional technologies, this invention offers numerous advantages. For example, the described NEMS oscillator network can be scaled up to very large networks (e.g., with hundreds of thousands of nodes) to enable extremely complex computations. In one example, a 1,000-node network can outperform current graphics processors by four orders of magnitude. As another example, a 100,000-node network can compete with, or surpass, large cloud-based systems / services currently used for machine learning. Each NEMS element in the system can directly generate RF electrical signals, which are then routed in complex network topologies constructed using printed circuit boards, integrated circuits, silicon-based printed circuit boards, and interpolators or application-specific integrated circuits (ASICs). Using time-domain and frequency-domain multiplexing, embodiments of the invention provide network architectures with at least 20,000 edges (for printed circuit boards) or 5,000,000 edges (for ASICs).

[0013] Furthermore, the described NEMS oscillator networks enable a variety of computational paradigms. These include adiabatic classical annealing, reservoir computation, Boltzmann machine processing, and several other machine learning and neuromorphic computations. The extremely high dimensionality of the described NEMS networks is advantageous because it enables the implementation of complex neural networks. The described NEMS networks can be scaled down to chip-level systems, such as in a system-on-a-chip (SoC) configuration. In some embodiments, these networks can provide ultra-low-power analog accelerators that can be integrated into a complete system as an SoC. These and other embodiments of this disclosure, along with their many advantages and features, are described in more detail below with reference to the accompanying drawings. Attached Figure Description

[0014] Figure 1An example NEMS oscillator according to an embodiment of the present invention is shown.

[0015] Figure 2 An example configuration of a plurality of NEMS oscillators according to an embodiment of the present invention is shown.

[0016] Figure 3 An example NEMS oscillator network according to an embodiment of the present invention is shown.

[0017] Figure 4 A training scheme for a NEMS oscillator network according to an embodiment of the present invention is shown.

[0018] Figure 5 An example method for operating a NEMS oscillator network according to an embodiment of the present invention is shown.

[0019] Figure 6 Another example method for training a NEMS oscillator network according to an embodiment of the present invention is shown.

[0020] Figure 7 A simplified computer system according to an embodiment of the present invention is shown. Detailed Implementation

[0021] The embodiments described herein provide a nanoelectromechanical system (NEMS) oscillator network and a method for using and training such a network. In some embodiments, a simulation computer composed of a multi-node simulation network based on a self-sustaining NEMS oscillator is provided. Using these computers and corresponding networks, simulation calculations of multiple combinatorial optimization problems and other neuromorphic calculations can be performed.

[0022] Regarding the various described embodiments, NEMS devices can be mechanical resonant elements made of semiconductors, piezoelectrics, metals, and other materials. A characteristic of such devices is that the element exhibits a strong and readily accessible nonlinearity. This nonlinearity includes characteristics such as the variation of resonant frequency and / or quality factor (damping rate) with the amplitude of vibration. For NEMS devices, readily accessible nonlinearity may mean a small-amplitude frequency shift of 1% to 10% at low excitation voltages from 1mV to 10V.

[0023] NEMS devices exhibit several types of nonlinearity. In some cases, there is a "duffing" nonlinearity, where the frequency shift is proportional to the square of the amplitude, and the frequency can rise or fall. In other cases, there is a nonlinearity associated with changes in damping rate, such that the damping rate increases with the amplitude and / or the square of the amplitude. This increase in damping leads to a decrease in frequency. In some cases, the duffing shift can be much larger than the damping shift.

[0024] In some embodiments, when an excitation input voltage is supplied to the NEMS resonator, the amplitude of the NEMS resonator changes, wherein the relationship between the excitation input voltage and the amplitude is linear or nonlinear. When the relationship between the amplitude and the excitation input voltage is nonlinear, the change in the amplitude of the NEMS resonator will, in turn, cause a change in the resonant frequency, dissipation, and / or phase-amplitude relationship of the NEMS resonator. By means of these relationships, each NEMS resonator is characterized by a nonlinear relationship between the excitation input voltage supplied to the NEMS resonator and the resonant frequency of the NEMS resonator. In some embodiments, nonlinear effects, such as changes in the resonant frequency, also affect the radio frequency (RF) output signal generated by the corresponding NEMS oscillator including the NEMS resonator (e.g., changes in frequency and / or amplitude). In a network of oscillator coupling, the output of one oscillator is sent to the input of another oscillator. Therefore, nonlinear effects allow the frequency of one oscillator to be changed by the frequency of another oscillator.

[0025] An example of an NEMS network receiving input signals of various kinds is described. As a result, the network oscillates and evolves with complex dynamics. This dynamics can be used for clustering and unsupervised learning in machine learning applications. The dynamics can project input signals (at the current time and previously received input signals) onto high-dimensional states of the dynamics. Such states can then be grouped and classified into different categories (“clustering”) for identification, classification, decision-making, and other applications.

[0026] In the following description, various examples will be described. Specific configurations and details are set forth for illustrative purposes to provide a thorough understanding of the examples. However, it will also be apparent to those skilled in the art that the examples can be implemented without specific details. Furthermore, well-known features may be omitted or simplified so as not to obscure the described embodiments.

[0027] Figure 1 An example NEMS oscillator 120 according to some embodiments of the present disclosure is shown. The NEMS oscillator 120 includes an input 102, an output 104, a NEMS resonator 110, and a gain element 108 positioned along a feedback path 106 extending between the output 104 and the input 102. The input 102 may include AC drive, DC bias, stress bias, and temperature bias, among other possibilities. The output 104 may include one or more of these components. As described herein, an RF output signal may be generated at the output 104. The RF output signal may oscillate at a specific frequency and a specific phase.

[0028] In some embodiments, when powered by feedback path 106 to oscillate at RF (20 kHz to 300 GHz), the NEMS oscillator 120 can be a non-linear oscillator. This can be achieved at least in part by the feedback path 106 that includes linear and parametric feedback components. In some examples, the NEMS oscillator 120 is driven non-linearly at very low voltages (1 to 10 mV) and very low powers (< nW each). This enables low-power, large-scale operation of the NEMS oscillator network.

[0029] In some embodiments, the NEMS resonator 110 can be directly driven by an AC driver. When implemented in a network, the NEMS oscillator 120 can generate an output RF signal at output 104 and send it into the network. The NEMS oscillator 120 can also receive signals from other components (e.g., other NEMS oscillators) in the network at input 102. The frequency of the NEMS oscillator 120 can be precisely tuned (resolution better than 1 ppm) through DC biasing, mechanical stress, temperature control, etc. Although the NEMS resonator 110 can act as a transceiver of RF electrical signals, in some embodiments, it can use different excitation and detection methods, including piezoelectric, thermal excitation, piezoresistive detection, etc.

[0030] The NEMS oscillator 120 can interact with other NEMS oscillators through its generated RF output signal. For example, the actual instantaneous frequency and phase of the RF output signal oscillation can vary depending on the state of the oscillation itself and the input of the network. In some embodiments, the rule of interaction can be that two oscillators are within the "line-width". For example, for a resonator with 10 MHz and quality factor Q = 1000, the line-width can be 10 MHz / 1000 = 10 kHz. Further descriptions of the NEMS oscillator and its operation can be found in U.S. Patent No. 9,660,654, titled "SYNCHRONIZATION OF NANOMECHANICAL OSCILLATORS", the disclosure of which is incorporated herein by reference in its entirety.

[0031] Figure 2 An example configuration of multiple NEMS oscillators 220 according to some embodiments of the present disclosure is shown. The illustrated configuration can correspond to a NEMS oscillator network integrated with a control and readout system, which can be implemented by a computing system 206 and one or more digital-to-analog converters (DACs) and analog-to-digital converters (ADCs). The computing system 206 can generate an input signal 244, which can correspond to the input signal of the NEMS oscillator network. The computing system 206 can also read an output signal 246, which can correspond to the output signal of the NEMS oscillator network.

[0032] NEMS oscillators 220 are interconnected via connections 222. For example, the output of NEMS oscillator 220-1 is connected to the input of NEMS oscillator 220-2 and the input of NEMS oscillator 220-3, and the output of NEMS oscillator 220-3 is connected to the input of NEMS oscillator 220-1. Note that NEMS oscillators 220-2 and 220-3 are not directly connected. In the illustrated example, each connection 222 includes a coupling weight 226 that affects the RF signal passing through it. For example, each coupling weight 226 can cause variations in the amplitude of the RF signal, variations in the phase of the RF signal, and other possibilities. In some embodiments, each coupling weight 226 included on the connection 222 between two NEMS oscillators 220 (e.g., between the output of the first NEMS oscillator and the input of the second NEMS oscillator) can allow only a small fraction of the RF output signal to pass through, for example, 1% or 2% of the RF signal amplitude.

[0033] The computing system 206 can generate a control signal 234, which is fed into the NEMS oscillator network to modify the function of the NEMS oscillator 220 and the connection 222 (via coupling weight 226). For example, the control signal 234 can be provided to the input of the NEMS oscillator 220, the input of the NEMS resonator 210, the gain of the NEMS oscillator 220, the coupling weight 226, and other possibilities. The control signal 234 provided to the coupling weight 226 can modify and / or set the amplitude and phase variations associated with the coupling weight 226.

[0034] One aspect of the NEMS oscillator network is that it is fully and rapidly configurable. In some embodiments, this means that connectivity is updated in real time. Another aspect is that the connections 222 within the NEMS oscillator network are fully analog. Fully analog networks allow for simultaneous parallel operation and low-power operation. In some embodiments, the control signal 234 can be analog or digital, while the connections 222 carry analog signals. In some embodiments, the control signal 234 can modify the coupling weights 226 to open or close the connection. In some embodiments, the control signal 234 can modify multiple configurable parameters, including: connection on / off, connection amplitude, connection phase, the inherent frequency of each NEMS oscillator, etc.

[0035] In some implementations, the network is composed of passive and active radio frequency (RF) components, such as resistors, transistors, and inductors. These networks can be characterized using the properties of conventional RF networks, such as transmission, reflection, isolation, and crosstalk between two paths (corresponding to two distinct edges and connections in the network). Electrical isolation methods, such as terminations and buffer amplifiers, are applied accordingly.

[0036] Figure 3An example NEMS oscillator network 300 according to some embodiments of the present disclosure is illustrated. The NEMS oscillator network 300 may include one or more network inputs 301 forming an input layer 302, a plurality of NEMS oscillators 320 interconnected via connections 322 to form a memory 324, and one or more network outputs 303 forming an output layer 304. In the illustrated example, network inputs 301 are connected via connections 322 to a first subset of the NEMS oscillators 320, and a second subset of the NEMS oscillators 320 are connected via connections 322 to the network outputs 303. Each connection 322 may include coupling weights 326 that modify RF signals transmitted therethrough. For example, each coupling weight 326 may allow a certain percentage of the output RF signal generated by one NEMS oscillator 320 to be sent to the input of another NEMS oscillator.

[0037] Each connection 322 can be unidirectional or bidirectional. A unidirectional connection between two NEMS oscillators can include a connection between the output of one NEMS oscillator and the input of the other. A bidirectional connection between two NEMS oscillators can include a first connection between the output of a first NEMS oscillator and the input of a second NEMS oscillator, and a second connection between the output of the second NEMS oscillator and the input of the first NEMS oscillator. Although Figure 3 A single coupling weight 326 is shown for each connection 322, but a bidirectional connection may include two different coupling weights for each direction. Typically, the connection 322 between the network input 301 and a first subset of the NEMS oscillators 320 can be a unidirectional connection, as can the connection 322 between a second subset of the NEMS oscillators 320 and the network output 303.

[0038] For illustrative purposes, the arrows for each NEMS oscillator 320 can indicate the phase of the RF output signal generated by the corresponding NEMS oscillator, and the rotation of the arrows (not shown) can indicate the frequency of the RF output signal generated by the corresponding NEMS oscillator. For example, two NEMS oscillators with a 180-degree phase difference can be represented by upward and downward arrows. As another example, two NEMS oscillators with a frequency difference can be described by a static arrow representing an oscillation at a nominal frequency (e.g., 2 MHz) and another slowly rotating clockwise or counterclockwise arrow representing an oscillation at a nominal frequency slightly higher (e.g., 2.001 MHz) or slightly lower (e.g., 1.999 MHz). The frequency and phase of the oscillation indicate the state of the oscillator. The state of the oscillator follows both its own dynamic characteristics and the inputs of other oscillators in the network. Generally, when two oscillators interact, their frequencies are so close that they are within the linewidth. In other cases, nonlinear mixing involving two, three, or four frequencies is possible when the mixing products are also within the interaction range.

[0039] During the operation or training of the NEMS oscillator network 300, one or more input signals 344 may be provided at one or more network inputs 301 of the input layer 302. For example, a first input signal 344 may be provided at a first network input 301, and a second input signal 344 may be provided at a second network input 301. Each input signal 344 may include multiple timing values, which, in some embodiments, may be converted into RF signals before the input signal 344 is provided at the network input 301.

[0040] Before, during, or after providing the input signal 344 at network input 301, each NEMS oscillator 320 may generate an RF output signal that oscillates at a specific frequency and a specific phase. This RF output signal may be referred to as the initial RF output signal. When the first timing value of the input signal 344 is provided at network input 301, the input signal is propagated through connection 322 to a first subset of the NEMS oscillators 320 coupled to network input 301. The RF output generated by each of the first subset of NEMS oscillators 320 is then adjusted to oscillate at a specific frequency (which may differ from the initial frequency) and a specific phase (which may differ from the initial phase) as a new RF output signal. This RF output signal may be referred to as the adjusted RF output signal.

[0041] The adjusted RF output signal generated by the first subset of NEMS oscillators 320 propagates through the remaining NEMS oscillators 320, causing each NEMS oscillator 320 to generate an adjusted RF output signal. The adjusted RF output signal generated by the second subset of NEMS oscillators 320 (oscillators coupled to network output 303) is fed to network output 303 and output by the NEMS oscillator network 300. These adjusted RF output signals can correspond to output signal 346, which is read by the computing system and used for calculations.

[0042] In some embodiments, the NEMS oscillator network 300 can be used for time series forecasting. In one example, the input signal 344 provided at each network input 301 can correspond to the stock prices of different companies, such as companies A, B, and C. The stock prices P of companies A, B, and C are... A (T), P B (T) and P C (T) can be input to the input layer 302 by modulating the amplitude of the AC drive, for example, V(t=Pa(t / T)×sin(ωt), where V(t) is the voltage at one of the network inputs 301. Because the resonator has a very fast response time, the stock price of an entire year can be compressed into a very short pulse (e.g., 0.01ms). The prices of different stocks are input to different nodes of the input layer 302 (e.g., network input 301).

[0043] Continuing the example above, stock prices from day 1 to day 100 might be available, and it might be desirable to predict the stock price on day 101. Stock prices from day 1 to day 10 (for company A only or for all companies) can be entered at input layer 302. The output V(10) is read at one of the network outputs 303 and compared with the actual price on day 11 = P. A (11) Compare. V(10) and P A The mapping between (11) can be developed through linear mapping based on reservoir calculations, thereby reconciling the prediction process with reality. This can correspond to the training steps used to train the NEMS oscillator network 300. The process can be repeated, for example, using data from day 2 to day 11 to predict day 12, using data from day 3 to day 12 to predict day 13, and so on, up to day 100.

[0044] Once the NEMS oscillator network 300 is trained, it can be used in a runtime scenario. Continuing the example above, stock price data from day 91 to day 100 can be input at input layer 302. The output V(100) is then processed by a linear mapping to predict P. A(101, Predicted). In some embodiments, the mapping may utilize two or more output signals 346 to predict P. A (101, predicted). For example, a 2D mapping can be used to map V1(100) and V2(100) to predict P. A (101, Predicted). Embodiments in which multiple input signals 344 and multiple output signals 346 are used for training and prediction are contemplated and are within the scope of this disclosure. Such prediction and forecasting networks can be applied to weather forecasting, sales forecasting, process model prediction and optimization. Health monitoring and event prediction include predicting cardiac arrest based on continuous observation and analysis of signals from the body, such as electroencephalograms and blood pressure. Prediction can also be used in scientific and engineering applications such as pattern formation, neurobiology, and chaotic systems.

[0045] In some embodiments, the NEMS oscillator network 300 supports the propagation and interaction of phase solitons. A phase soliton refers to a short-term, minute change in the oscillator frequency. This change can move non-dissipatively from one oscillator to another in the network (the change does not diminish or spatially change shape). When two solitons collide / meet, they interact and can change their amplitude and shape. This interaction can be used for computation. The NEMS soliton system supports all basic function computations, including cascading capabilities (where the output of one part becomes the input of another), fan-out (connecting to multiple outputs), and gate operation performance, such as NOT gates and other two-input gates.

[0046] Figure 4 Training schemes for NEMS oscillator networks according to some embodiments of the present disclosure are illustrated. Training data 450 can be provided to computing system 406. Training data 450 may include one or more input signals and one or more reference output signals. The reference output signals may be known or desired outputs corresponding to the input signals. In one example, the input signals may be the stock prices of three companies from day 1 to day 10, and the output signals may be the stock prices of the three companies (or one of the three companies) on day 11.

[0047] When training data 450 is provided to computing system 406, the NEMS oscillator network can operate using the input signal. The output signal read by computing system 406 can be compared with a reference output signal to generate error data representing the difference between the output signal and the reference output signal. Computing system 406 can then modify the NEMS oscillator network based on the error data. For example, computing system 406 can generate a control signal 434 that modifies one or more NEMS oscillators 420 or one or more coupling weights 426. Computing system 406 can alternatively or additionally modify the mapping 470 between the output signal and some value of interest (e.g., stock price). In some embodiments, the NEMS oscillator network can be modified such that the prediction error is reduced (as represented by the error data) during subsequent predictions using the same input data. In this way, the NEMS oscillator network can be trained to accurately generate output data for a wide range of input data.

[0048] Figure 5 Example methods 500 for operating a NEMS oscillator network (e.g., NEMS oscillator network 300) according to some embodiments of the present disclosure are illustrated. One or more steps of method 500 may be omitted during execution of method 500, and the steps of method 500 need not be performed in the order shown. One or more steps of method 500 may be performed by one or more processors, such as processors included in a computing system (e.g., computing systems 206, 406). Method 500 may be implemented as a computer-readable medium or computer program product including instructions that, when executed by one or more computers, cause the one or more computers to perform the steps of method 500. Such a computer program product may be transmitted in a data carrier signal carrying the computer program product via a wired or wireless network.

[0049] In step 502, one or more input signals (e.g., input signals 244, 344) are received at one or more network inputs (e.g., network input 301) of the NEMS oscillator network. The one or more network inputs may form an input layer (e.g., input layer 302). The NEMS oscillator network may include multiple NEMS oscillators (e.g., NEMS oscillators 120, 220, 320, 420). In some embodiments, each of the multiple NEMS oscillators includes a NEMS resonator (e.g., NEMS resonators 110, 210). In some embodiments, each of the multiple NEMS oscillators includes a feedback path (e.g., feedback 106) between the output (e.g., output 104) and the input (e.g., input 102) of the NEMS resonator. In some embodiments, there is a nonlinear relationship between the excitation input voltage provided to the NEMS resonator and the amplitude of the RF output signal, the phase of the RF output signal, and / or the resonant frequency of the NEMS resonator.

[0050] In step 504, an RF output signal is generated at each of the plurality of NEMS oscillators. The RF output signal may oscillate at a specific frequency and a specific phase. The plurality of NEMS oscillators may be coupled to one or more network inputs. The NEMS oscillator network may include multiple connections (e.g., connections 222, 322) interconnecting the plurality of NEMS oscillators. In some examples, each of the plurality of NEMS oscillators is connected to at least one other NEMS oscillator among the plurality of NEMS oscillators via at least one of the multiple connections.

[0051] In some embodiments, each of the plurality of connections includes a connection weight (e.g., connection weights 226, 326, 426). In some embodiments, the coupling weights can cause amplitude variations in the RF output (or input) signal through the connection including the coupling weight. In some embodiments, the coupling weights can cause phase variations in the RF output (or input) signal through the connection including the coupling weight. In some embodiments, the RF output signal generated at each of the plurality of NEMS oscillators is transmitted through at least one of the plurality of connections.

[0052] In some embodiments, step 504 may include one or both of steps 506 and 508. In step 506, an initial RF output signal is generated at each of the plurality of NEMS oscillators. In step 508, an updated RF output signal is generated at each of the plurality of NEMS oscillators. The updated RF output signal may be generated in response to one or more input signals provided at one or more network inputs.

[0053] In step 510, one or more output signals (e.g., output signals 246, 346) are output at one or more network outputs of the NEMS oscillator network (e.g., network output 303). The one or more network outputs may be coupled to multiple NEMS oscillators. The one or more network outputs may form an output layer (e.g., output layer 304). The computing system can read the one or more output signals. A mapping (e.g., mapping 470) can be used to convert the one or more output signals into values ​​of interest.

[0054] Figure 6 An example method 600 for training a NEMS oscillator network (e.g., NEMS oscillator network 300) according to some embodiments of the present disclosure is illustrated. One or more steps of method 600 may be omitted during execution of method 600, and the steps of method 600 do not need to be performed in the order shown. One or more steps of method 600 may be performed by one or more processors, such as processors included in a computing system (e.g., computing systems 206, 406). Method 600 may be implemented as a computer-readable medium or computer program product including instructions that, when executed by one or more computers, cause the one or more computers to perform the steps of method 600. Such a computer program product may be transmitted in a data carrier signal carrying the computer program product via a wired or wireless network.

[0055] In step 602, one or more input signals (e.g., input signals 244, 344) are provided at one or more network inputs (e.g., network input 301) of the NEMS oscillator network. The one or more network inputs may form an input layer (e.g., input layer 302). The NEMS oscillator network may include multiple NEMS oscillators (e.g., NEMS oscillators 120, 220, 320, 420). In some embodiments, each of the multiple NEMS oscillators includes a NEMS resonator (e.g., NEMS resonators 110, 210). In some embodiments, each of the multiple NEMS oscillators includes a feedback path (e.g., feedback 106) between the output (e.g., output 104) and the input (e.g., input 102) of the NEMS resonator.

[0056] In step 604, an RF output signal is generated at each of the plurality of NEMS oscillators. The RF output signal may oscillate at a specific frequency and a specific phase. The plurality of NEMS oscillators may be coupled to one or more network inputs. The NEMS oscillator network may include multiple connections (e.g., connections 222, 322) interconnecting the plurality of NEMS oscillators. In some examples, each of the plurality of NEMS oscillators is connected to at least one other NEMS oscillator among the plurality of NEMS oscillators via at least one of the multiple connections.

[0057] In some embodiments, each of the plurality of connections includes a plurality of connection weights (e.g., connection weights 226, 326, 426). In some embodiments, the coupling weights may cause amplitude variations in the RF output signal (or input signal) through the connection including the coupling weights. In some embodiments, the coupling weights may cause phase variations in the RF output signal (or input signal) through the connection including the coupling weights. In some embodiments, the RF output signal generated at each of the plurality of NEMS oscillators is transmitted through at least one of the plurality of connections.

[0058] In some embodiments, step 604 may include one or both of steps 606 and 608. In step 606, an initial RF output signal is generated at each of the plurality of NEMS oscillators. In step 608, an updated RF output signal is generated at each of the plurality of NEMS oscillators. The updated RF output signal may be generated in response to one or more input signals provided at one or more network inputs.

[0059] In step 610, one or more output signals (e.g., output signals 246, 346) are read from one or more network outputs of the NEMS oscillator network (e.g., network output 303). The one or more network outputs may be coupled to multiple NEMS oscillators. The one or more network outputs may form an output layer (e.g., output layer 304).

[0060] In step 612, a comparison may be performed between one or more output signals and one or more reference output signals. In some embodiments, error data representing the difference between the one or more output signals and the one or more reference output signals may be calculated.

[0061] In step 614, the NEMS oscillator network can be modified based on comparison. In some embodiments, step 614 may include one or both of steps 616 and 618. In step 616, the amplitude or phase change of at least one of the plurality of coupling weights is modified based on comparison. In step 618, a specific frequency of at least one of the plurality of NEMS oscillators is modified based on comparison. In some embodiments, modifying the NEMS oscillator network based on comparison results in a reduction in the error / difference between one or more output signals and one or more reference output signals. In some embodiments, steps 602 through 618 may be repeated for each of the plurality of training iterations.

[0062] Figure 7 A simplified computer system 700 according to some embodiments of the present disclosure is shown. For example... Figure 7 The computer system 700 shown can be incorporated into the device described herein. Figure 7 A schematic diagram of one embodiment of a computer system 700 is provided, which can perform some or all of the steps of the methods provided by various embodiments. It should be noted that... Figure 7 This is intended only to provide a general overview of the various components; any one or all of them may be used appropriately. Therefore, Figure 7 It outlines how individual system components can be implemented in a relatively separate or relatively more integrated manner.

[0063] Computer system 700 is shown to include hardware elements that may be electrically coupled via bus 705 or otherwise appropriately communicated. The hardware elements may include: one or more processors 710, including but not limited to one or more general-purpose processors and / or one or more special-purpose processors, such as digital signal processing chips, graphics accelerators, etc.; one or more input devices 715, which may include but are not limited to mice, keyboards, cameras, etc.; and one or more output devices 720, which may include but are not limited to display devices, printers, etc.

[0064] The computer system 700 may also include and / or communicate with one or more non-transitory storage devices 725, which may include, but are not limited to, local and / or network-accessible storage, and / or may include, but are not limited to, disk drives, drive arrays, optical storage devices, solid-state storage devices, such as random access memory (“RAM”) and / or read-only memory (“ROM”), which may be programmable, flash-updatable, etc. Such storage devices can be configured to implement any suitable data storage, including but not limited to various file systems, database structures, etc.

[0065] Computer system 700 may also include a communication subsystem 719, which may include, but is not limited to, a modem, a network interface card (NIC) (wireless or wired), an infrared communication device, a wireless communication device, and / or a chipset, such as a Bluetooth™ device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication facility, etc. Communication subsystem 719 may include one or more input and / or output communication interfaces to allow data exchange with a network (such as the network described below as an example), other computer systems, a television, and / or any other device described herein. Depending on desired functionality and / or other implementation considerations, portable electronic devices or similar devices may communicate images and / or other information via communication subsystem 719. In other embodiments, portable electronic devices, such as a first electronic device, may be incorporated into computer system 700, for example, as an electronic device serving as input device 715. In some embodiments, computer system 700 will also include working memory 735, which, as described above, may include RAM or ROM devices.

[0066] Computer system 700 may also include software elements, shown as currently residing within working memory 735, including operating system 740, device drivers, executable libraries, and / or other code, such as one or more application programs 745, which may include computer programs provided by various embodiments and / or may be designed to implement methods and / or configure systems provided by other embodiments, as described herein. By way of example only, one or more processes described with respect to the above methods may be implemented as code and / or instructions executable by a computer and / or a processor within a computer; in one aspect, such code and / or instructions may be used to configure and / or adapt a general-purpose computer or other device to perform one or more operations according to the described methods.

[0067] A set of these instructions and / or code may be stored on a non-transitory computer-readable storage medium, such as the storage device 725 described above. In some cases, the storage medium may be incorporated into a computer system, such as computer system 700. In other embodiments, the storage medium may be separate from the computer system (e.g., a removable medium such as an optical disc) and / or provided in an installation package, such that the storage medium can be used to program, configure, and / or adapt to a general-purpose computer on which the instructions / code are stored. These instructions may take the form of executable code, which can be executed by computer system 700, and / or may take the form of source code and / or installable code, for example, when compiled and / or installed on computer system 700, using any of a variety of generally available compilers, installers, compression / decompression utilities, etc., and then in the form of executable code.

[0068] It will be apparent to those skilled in the art that substantial modifications can be made to suit specific requirements. For example, custom hardware may be used, and / or specific components may be implemented in hardware, software including portable software (such as applets), or both. Furthermore, connections to other computing devices, such as network input / output devices, may be employed.

[0069] As described above, in one aspect, some embodiments may employ a computer system, such as computer system 700, to perform methods according to various embodiments of the present technology. According to one set of embodiments, some or all of the processes of these methods are executed by computer system 700 in response to processor 710 executing one or more sequences of one or more instructions, which may be incorporated into operating system 740 and / or other code (such as application program 745) contained in working memory 735. Such instructions may be read into working memory 735 from another computer-readable medium, such as one or more storage devices 725. By way of example only, execution of a sequence of instructions contained in working memory 735 may cause processor 710 to perform one or more processes of the methods described herein. Additionally or alternatively, portions of the methods described herein may be executed by dedicated hardware.

[0070] As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any medium that participates in providing data that enables a machine to operate in a particular manner. In embodiments implemented using computer system 700, various computer-readable media may participate in providing instructions / code to processor 710 for execution and / or may be used to store and / or carry such instructions / code. In many embodiments, the computer-readable medium is a physical and / or tangible storage medium. Such a medium may take the form of a non-volatile medium or a volatile medium. Non-volatile media include, for example, optical discs and / or magnetic disks, such as storage device 725. Volatile media include, but are not limited to, dynamic memory, such as working memory 735.

[0071] Common forms of physical and / or tangible computer-readable media include, for example, floppy disks, hard disks, magnetic tape or any other magnetic media, optical discs, any other optical media, punched cards, paper tape, any other physical media with a perforated pattern, random access memory, programmable read-only memory, flash memory, any other memory chip or cassette tape, or any other medium from which a computer may read instructions and / or code.

[0072] Various forms of computer-readable media may involve transmitting one or more sequences of one or more instructions to processor 710 for execution. By way of example only, the instructions may initially be carried on a disk and / or optical disk of a remote computer. The remote computer may load the instructions into its dynamic memory and transmit the instructions as signals via a transmission medium for reception and / or execution by computer system 700.

[0073] The communication subsystem 719 and / or its components typically receive signals, and then bus 705 can carry signals and / or data, instructions, etc. These signals are then carried to working memory 735, from which processor 710 retrieves and executes instructions. Instructions received by working memory 735 may optionally be stored on non-transitory storage device 725 before or after execution by processor 710.

[0074] The methods, systems, and apparatus discussed above are examples. Various configurations may appropriately omit, substitute, or add various procedures or components. For example, in alternative configurations, methods may be performed in a different order than described, and / or various stages may be added, omitted, and / or combined. Furthermore, features described with respect to certain configurations may be combined in various other configurations. Different aspects and elements of a configuration may be combined in a similar manner. Moreover, technology is evolving; therefore, many elements are examples and do not limit the scope of this disclosure or the claims.

[0075] Specific details are set forth in the description to provide a thorough understanding of the exemplary configurations, including the implementation. However, the configurations may be implemented without these specific details. For example, well-known circuits, processes, algorithms, structures, and techniques have been shown without unnecessary detail to avoid obscuring the configuration. This description provides only exemplary configurations and does not limit the scope, applicability, or configuration of the claims. Rather, the foregoing description of the configurations will provide those skilled in the art with an enabling description for implementing the techniques described. Various changes may be made to the function and arrangement of the elements without departing from the spirit or scope of this disclosure.

[0076] Furthermore, the configuration can be described as a process depicted as a schematic flowchart or block diagram. Although each operation can be described as a sequential process, many operations can be executed in parallel or concurrently. Moreover, the order of operations can be rearranged. A process may have additional steps not included in the diagram. Furthermore, examples of the method can be implemented using hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware, or microcode, the program code or code segments that perform the necessary tasks can be stored in a non-transitory computer-readable medium such as a storage medium. The processor can execute the described tasks.

[0077] Several example configurations have been described, and various modifications, alternative constructions, and equivalents may be used without departing from the spirit of this disclosure. For example, the aforementioned elements may be components of a larger system in which other rules may take precedence over or otherwise modify the application of the technology. Furthermore, numerous steps may be taken before, during, or after considering the aforementioned elements. Therefore, the above description does not limit the scope of the claims.

[0078] As used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly specifies otherwise. Thus, for example, a reference to “user” includes a plurality of such users, a reference to “processor” includes a reference to one or more processors and their equivalents known to those skilled in the art, and so on.

[0079] Furthermore, when used in this specification and the following claims, the words “comprising,” “including,” “comprise,” “include,” “include,” and “include” are intended to specify the presence of the stated features, integrals, components, or steps, but they do not exclude the presence or addition of one or more other features, integrals, components, steps, actions, or groups.

[0080] It should also be understood that the examples and embodiments described herein are for illustrative purposes only, and various modifications or variations based on these examples and embodiments will be suggested to those skilled in the art and will be included within the spirit and scope of this application and the scope of the appended claims.

Claims

1. A nano-electromechanical system (NEMS) oscillator network comprising: one or more network inputs configured to receive one or more input signals; a plurality of NEMS oscillators coupled to the one or more network inputs, wherein each NEMS oscillator of the plurality of NEMS oscillators produces a radio frequency (RF) output signal oscillating at a particular frequency and a particular phase; a plurality of connections interconnecting the plurality of NEMS oscillators such that each NEMS oscillator of the plurality of NEMS oscillators is connected to at least one other NEMS oscillator of the plurality of NEMS oscillators through at least one connection of the plurality of connections, wherein outputs of a first portion of the plurality of NEMS oscillators are connected to inputs of a second portion of the plurality of NEMS oscillators; a plurality of coupling weights included in the plurality of connections, wherein each of the plurality of coupling weights is operable to cause a change in amplitude of the RF output signal through a respective connection of the plurality of connections, wherein the plurality of coupling weights are modified during a training process; and one or more network outputs coupled to the plurality of NEMS oscillators, wherein the one or more network outputs are configured to output one or more output signals. each NEMS oscillator of the plurality of NEMS oscillators comprises a NEMS resonator, wherein there is a non-linear relationship between an excitation input voltage provided to the NEMS resonator and an amplitude and a phase of the RF output signal.

2. The NEMS oscillator network of claim 1, wherein, each NEMS oscillator of the plurality of NEMS oscillators further comprises a feedback path between an output of the NEMS resonator and an input of the NEMS resonator.

3. The NEMS oscillator network of claim 2, wherein, at least one of the plurality of coupling weights is located between the one or more network inputs and the plurality of NEMS oscillators.

4. The NEMS oscillator network of claim 1, wherein, each of the plurality of coupling weights is further operable to cause a change in phase of the RF output signal through a respective connection.

5. The NEMS oscillator network of claim 1, wherein, the RF output signal produced at each NEMS oscillator of the plurality of NEMS oscillators is operable to be transmitted through at least one connection of the plurality of connections.

6. The NEMS oscillator network of claim 1, wherein, 7. The NEMS oscillator network of claim 1, further comprising a processor configured to perform operations of a training process, the operations comprising: providing the one or more input signals at the one or more network inputs; reading the one or more output signals at the one or more network outputs; performing a comparison between the one or more output signals and one or more reference output signals; and modifying the NEMS oscillator network based on the comparison.

8. A method of operating a nano-electromechanical system (NEMS) oscillator network, the method comprising: receiving one or more input signals at one or more network inputs of the NEMS oscillator network; ​ ​ producing, for each of a plurality of NEMS oscillators of the NEMS oscillator network, a radio frequency, RF, output signal oscillating at a particular frequency and a particular phase, wherein the plurality of NEMS oscillators are coupled to the one or more network inputs, and wherein the NEMS oscillator network comprises a plurality of connections interconnecting the plurality of NEMS oscillators such that each of the plurality of NEMS oscillators is connected to at least one other of the plurality of NEMS oscillators by at least one of the plurality of connections, wherein an output of a first portion of the plurality of NEMS oscillators is connected to an input of a second portion of the plurality of NEMS oscillators, wherein the plurality of connections comprises a plurality of coupling weights, and wherein each of the plurality of coupling weights is operable to cause a change in amplitude of the RF output signal through a respective one of the plurality of connections, wherein the plurality of coupling weights are modified during a training process; and outputting one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs being coupled to the plurality of NEMS oscillators.

9. The method of claim 8, wherein, each of the plurality of NEMS oscillators comprises a NEMS resonator, wherein there is a non-linear relationship between an excitation input voltage provided to the NEMS resonator and an amplitude and a phase of the RF output signal.

10. The method of claim 9, wherein, each of the plurality of NEMS oscillators further comprises a feedback path between an output of the NEMS resonator and an input of the NEMS resonator.

11. The method of claim 8, wherein, at least one of the plurality of coupling weights is located between the one or more network inputs and the plurality of NEMS oscillators.

12. The method of claim 8, wherein, each of the plurality of coupling weights further causes a change in phase of the RF output signal through the respective connection.

13. The method of claim 8, further comprising: sending, through at least one of the plurality of connections, the RF output signal produced at each of the plurality of NEMS oscillators.

14. A method of training a network of nano-electromechanical system, NEMS, oscillators, the method comprising: providing one or more input signals to one or more network inputs of the NEMS oscillator network; producing, for each of a plurality of NEMS oscillators of the NEMS oscillator network, a radio frequency, RF, output signal oscillating at a particular frequency and a particular phase, wherein the plurality of NEMS oscillators are coupled to the one or more network inputs, and wherein the NEMS oscillator network comprises a plurality of connections interconnecting the plurality of NEMS oscillators such that each of the plurality of NEMS oscillators is connected to at least one other of the plurality of NEMS oscillators by at least one of the plurality of connections, wherein a plurality of coupling weights are included in the plurality of connections, wherein each of the plurality of coupling weights is operable to cause a change in amplitude of the RF output signal through a respective one of the plurality of connections, and wherein an output of a first portion of the plurality of NEMS oscillators is connected to an input of a second portion of the plurality of NEMS oscillators; reading one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs being coupled to the plurality of NEMS oscillators; performing a comparison between the one or more output signals and the one or more reference output signals; and modifying the NEMS oscillator network based on the comparison. each of the plurality of NEMS oscillators comprises a NEMS resonator, wherein there is a non-linear relationship between an excitation input voltage provided to the NEMS resonator and an amplitude and a phase of the RF output signal.

15. The method of claim 14, wherein, each of the plurality of NEMS oscillators further comprises a feedback path between an output of the NEMS resonator and an input of the NEMS resonator.

16. The method of claim 15, wherein, each of the plurality of connections comprises a coupling weight of the plurality of coupling weights.

17. The method of claim 14, wherein, modifying the NEMS oscillator network based on the comparison comprises modifying a change in amplitude for at least one of the plurality of coupling weights.

18. The method of claim 17, wherein, each of the plurality of coupling weights further causes a change in phase of the RF output signal through the connection.

19. The method of claim 17, wherein, 20. The method of claim 14, further comprising: sending, through at least one of the plurality of connections, the RF output signal produced at each of the plurality of NEMS oscillators.

21. A method of training a network of nano-electromechanical system, NEMS, oscillators, the method comprising: providing, from a training data set, one or more input signals to one or more network inputs of the NEMS oscillator network; ​ for each of a plurality of NEMS oscillators of the NEMS oscillator network, generating a radio frequency (RF) output signal oscillating at a particular frequency and a particular phase, wherein the plurality of NEMS oscillators are coupled to the one or more network inputs, and wherein the NEMS oscillator network comprises a plurality of connections interconnecting the plurality of NEMS oscillators such that each of the plurality of NEMS oscillators is connected to at least one other of the plurality of NEMS oscillators by at least one of the plurality of connections, wherein a plurality of coupling weights are included in the plurality of connections, wherein each of the plurality of coupling weights is operable to cause a change in amplitude of the RF output signal through a respective one of the plurality of connections, and wherein an output of a first portion of the plurality of NEMS oscillators is connected to an input of a second portion of the plurality of NEMS oscillators; reading one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs being coupled to the plurality of NEMS oscillators; and modifying the NEMS oscillator network based on the one or more output signals.

22. The method of claim 21, further comprising: performing a comparison between the one or more output signals and one or more reference output signals from the training data set, wherein modifying the NEMS oscillator network based on the one or more output signals comprises modifying the NEMS oscillator network based on the comparison.

23. The method of claim 21, wherein, each of the plurality of NEMS oscillators comprises a NEMS resonator, wherein there is a non-linear relationship between an excitation input voltage provided to the NEMS resonator and an amplitude and a phase of the RF output signal.

24. The method of claim 23, wherein, each of the plurality of NEMS oscillators further comprises a feedback path between an output of the NEMS resonator and an input of the NEMS resonator.

25. The method of claim 21, wherein, each of the plurality of connections comprises a coupling weight of the plurality of coupling weights.

26. The method of claim 25, wherein, modifying the NEMS oscillator network comprises modifying a change in amplitude for at least one of the plurality of coupling weights.

27. The method of claim 25, wherein, each of the plurality of coupling weights further causes a change in phase of the RF output signal through the connection.

28. The method of claim 21, further comprising: sending, through at least one of the plurality of connections, the RF output signal generated at each of the plurality of NEMS oscillators.

29. A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations for training a nano-electromechanical system (NEMS) oscillator network, the operations comprising: providing one or more input signals from a training data set to one or more network inputs of the NEMS oscillator network; for each of a plurality of NEMS oscillators of the NEMS oscillator network, causing a radio frequency (RF) output signal to oscillate at a particular frequency and a particular phase, wherein the plurality of NEMS oscillators are coupled to the one or more network inputs, and wherein the NEMS oscillator network comprises a plurality of connections interconnecting the plurality of NEMS oscillators such that each of the plurality of NEMS oscillators is connected to at least one other of the plurality of NEMS oscillators by at least one of the plurality of connections, wherein a plurality of coupling weights are included in the plurality of connections, wherein each of the plurality of coupling weights is operable to cause a change in amplitude of the RF output signal through a respective one of the plurality of connections, and wherein an output of a first portion of the plurality of NEMS oscillators is connected to an input of a second portion of the plurality of NEMS oscillators; reading one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs coupled to the plurality of NEMS oscillators; and modifying the NEMS oscillator network based on the one or more output signals.

30. The non-transitory computer-readable medium of claim 29, further comprising: performing a comparison between the one or more output signals and one or more reference output signals from the training data set, wherein modifying the NEMS oscillator network based on the one or more output signals comprises modifying the NEMS oscillator network based on the comparison.

31. The non-transitory computer-readable medium of claim 29, wherein, each of the plurality of NEMS oscillators comprises a NEMS resonator, wherein there is a non-linear relationship between an excitation input voltage provided to the NEMS resonator and an amplitude and a phase of the RF output signal.

32. The non-transitory computer-readable medium of claim 31, wherein, each of the plurality of NEMS oscillators further comprises a feedback path between an output of the NEMS resonator and an input of the NEMS resonator.

33. The non-transitory computer-readable medium of claim 29, wherein, each of the plurality of connections comprises a coupling weight of the plurality of coupling weights.

34. The non-transitory computer-readable medium of claim 33, wherein, modifying the NEMS oscillator network comprises modifying a change in amplitude for at least one of the plurality of coupling weights.

35. The non-transitory computer-readable medium of claim 33, wherein, each of the plurality of coupling weights further causes a change in phase of the RF output signal through the connection.

36. A method of training a nano-electromechanical system (NEMS) oscillator network, the method comprising: receiving a training data set; and for each of a plurality of training iterations: providing one or more input signals from the training data set to one or more network inputs of the NEMS oscillator network; reading one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs coupled to the plurality of NEMS oscillators; and modifying the NEMS oscillator network based on the one or more output signals. for each of a plurality of NEMS oscillators of the NEMS oscillator network, generating a radio frequency (RF) output signal oscillating at a particular frequency and a particular phase, wherein the plurality of NEMS oscillators are coupled to the one or more network inputs, and wherein the NEMS oscillator network comprises a plurality of connections interconnecting the plurality of NEMS oscillators such that each of the plurality of NEMS oscillators is connected to at least one other of the plurality of NEMS oscillators by at least one of the plurality of connections, wherein a plurality of coupling weights are included in the plurality of connections, wherein each of the plurality of coupling weights is operable to cause a change in amplitude of the RF output signal through a respective one of the plurality of connections, and wherein an output of a first portion of the plurality of NEMS oscillators is connected to an input of a second portion of the plurality of NEMS oscillators; reading one or more output signals at one or more network outputs of the NEMS oscillator network, the one or more network outputs being coupled to the plurality of NEMS oscillators; and modifying the NEMS oscillator network based on the one or more output signals.

37. The method of claim 36, further comprising: for each of a plurality of training iterations: performing a comparison between the one or more output signals and one or more reference output signals from the training data set, wherein modifying the NEMS oscillator network based on the one or more output signals comprises modifying the NEMS oscillator network based on the comparison.

38. The method of claim 36, wherein, each of the plurality of NEMS oscillators comprises a NEMS resonator, wherein there is a non-linear relationship between an excitation input voltage provided to the NEMS resonator and an amplitude and a phase of the RF output signal.

39. The method of claim 36, wherein, each of the plurality of connections comprises a coupling weight of the plurality of coupling weights.

40. The method of claim 39, wherein, modifying the NEMS oscillator network comprises modifying a change in amplitude for at least one of the plurality of coupling weights.

Images