Reconfigurable event driven hardware using reservoir computing for monitoring an electronic sensor and waking a processor
Inventors
Lipasti, Mikko H. • Hashmi, Atif G. • Nere, Andrew • Tononi, Giulio
Assignees
National Science Foundation NSF • Wisconsin Alumni Research Foundation
Publication Number
US-10013048-B2
Publication Date
2018-07-03
Expiration Date
2033-01-25
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
The present inventors have recognized that proper utilization of reconfigurable event driven hardware may achieve optimum power conservation in energy constrained environments including a low power general purpose primary processor and one or more electronic sensors. Aspects of neurobiology and neuroscience, for example, may be utilized to provide such reconfigurable event driven hardware, thereby achieving energy-efficient continuous sensing and signature reporting in conjunction with the one or more electronic sensors while the primary processor enters a low power consumption mode. Such hardware is event driven and operates with extremely low energy requirements.
Core Innovation
The invention provides a system and method for energy-efficient continuous sensing and detection of trigger signatures from sensory streams in energy constrained environments. This is achieved by using reconfigurable event driven hardware, which operates with extremely low energy requirements, to continuously monitor one or more electronic sensors and detect spatial or temporal trigger signatures that indicate events of interest. When such trigger signatures are identified, the reconfigurable event driven hardware wakes the low-power primary processor to handle further processing.
The problem addressed is that continuous sensing applications today rely on a low power general purpose primary processor that continuously monitors sensor data streams to detect trigger signatures, leading to significant energy consumption. Despite existing solutions involving low-power microcontrollers for continuous sensing, these add overhead in terms of energy, space, and compute resources. The invention solves this by providing reconfigurable event driven hardware inspired by neurobiology and neuroscience principles, such as reservoir computing, which offloads sampling and signature detection tasks from the primary processor. This allows the primary processor to remain in a low power mode for extended periods, significantly reducing overall power dissipation in continuous sensing systems.
The reconfigurable event driven hardware may be implemented with a neural network structure, including components such as an Echo State Network and a Multi-Layered Perceptron Network, comprising digitally implemented leaky integrate-and-fire neurons. This hardware interfaces directly with sensors, capturing temporal behaviors and classifying trigger signatures without invoking the primary processor until necessary. It can be reconfigured during runtime to adapt to different sensing applications and trigger signature requirements, supporting monitoring of multiple sensors and various sensory modalities to meet dynamic application needs.
Claims Coverage
The claims cover key inventive features in methods and systems employing reconfigurable event driven hardware with reservoir computing for low-power sensory stream analysis and trigger signature detection.
Reconfigurable event driven hardware with reservoir computing
The hardware includes a capturing element configured to capture temporal behavior of sensory streams, and a classifying element configured to classify outputs to identify trigger signatures. These elements are constructed as neuron-based structures with configurable neuron parameters interconnected by a reconfigurable interconnect.
Energy-efficient sensory stream monitoring and processor management
The processor is placed into a low power consumption mode while the reconfigurable event driven hardware continuously monitors sensory streams. Upon detecting trigger signatures, the hardware communicates details to the processor and invokes the corresponding applications, thus conserving energy.
Runtime reconfiguration of trigger signature detection
The reconfigurable event driven hardware can be reconfigured during runtime of the application to detect different trigger signatures in response to application-level changes, allowing dynamic adaptability.
Use of specific reservoir computing architectures
The reservoir computing comprises a Liquid State Machine made of leaky-integrate-and-fire neurons for capturing temporal features, and a Multi-Layered Perceptron Network of similar neurons for classification, where the capturing element is implemented using the LSM and the classifying element using the MLPN.
Integrated system arrangements
The system comprises multiple electronic sensors, a processor with a low power mode, and reconfigurable event driven hardware interfaced with both. The hardware and sensors may be co-located in a first enclosure with the processor either in the same or a separate enclosure, enhancing deployment flexibility.
Together, these features provide a novel, reconfigurable neural network-based hardware solution that offloads sensing and detection tasks from the primary processor, enabling adaptive, energy-efficient, continuous monitoring and triggering in constrained devices.
Stated Advantages
Significant energy savings as the primary processor can remain in low power mode while the reconfigurable hardware continuously monitors sensors.
Low power consumption achieved by event driven operation, where energy is dissipated primarily upon sensor data arrival, without the overhead of dedicated microcontrollers.
Flexibility to monitor multiple sensors and detect a wide variety of spatial or temporal trigger signatures, supporting diverse and changing application needs.
Runtime reconfigurability allowing adaptation to new or different trigger signatures dynamically during application execution.
Documented Applications
Medical and health monitoring applications including EKG, EEG, pulse, blood pressure, and patient activity monitoring.
Environmental monitoring such as emission levels, pollutant concentrations, or seismic data.
Mobile consumer devices including smartphones and tablets using sensors like accelerometers to detect user activity or context-specific events such as entering a driver seat.
Interested in licensing this patent?