The ULP computing platform considered in the project has in general already good performances in terms of power consumption and support of basic use cases. In order to reach this goal the platform has been reduced in terms of memory availability and operating frequency. In order to implement more advanced use cases and applications, and to improve the quality of the actual supported applications new accelerated functions should be added to the system and new approaches for the computer vision algorithms should be considered.
The approach of convolutional neural networks will allow, with a limited increase in the used resources, the implementation of a variety of new applications and improve the identification of classes of objects, that will be the basis of any video surveillance application where the recognition of the actors on the scene is fundamental to automatically understand the actions happening in the scene itself.
OPERA will investigate the definition of new HW/SW methodologies for specific computer vision application with a new approach based on convolutional neural networks that will be implemented in the newly integrated HW/SW local processing nodes. For validating the results of the project OPERA realize a prototypes of smart local sensors build with a 3D integration of imaging sensors and processing cores, typically realized in different SoC technologies and miniaturizations, that cannot be merged with a classical approach, but should be connected through a 3D stacking realizing an hybrid SoC more efficient an optimized in terms of power consumption.
In order to obtain ultra-low power architecture, particular attention should be paid to the electrical design and the power optimization of every component which will contribute to the whole system behaviour. In fact, energy efficiency in new generation datacentres is normally focused on the power optimization of the data processing modules and the adopted wireless communication protocols. To reach a higher level of energy efficiency, we must also take into account the selection of the active components by giving priority to the techniques and the technologies that guarantees the optimum trade-off between system performance and power consumption constraints. As a practical example, take the case of the reconfigurable antenna which is part of the wireless transmitting system may be driven by means of Radio Frequency Microelectromechanical system (RF MEMS). These components embed a series of sub-millimetre-sized parts that act as a micro-mechanical radiofrequency switch. These parts are moved by electrostatic charge, requiring therefore very low power consumption during commutation and almost zero power consumption in static conditions.
It is noteworthy that, as reported in the “STRATEGIC RESEARCH AGENDA of EPoSS – THE EUROPEAN TECHNOLOGY PLATFORM ON SMART SYSTEMS INTEGRATION” 4 , the use of steerable antenna arrays will be important for the wireless connection between smart objects.
Regarding the RF communication OPERA wants to move the efforts on the optimization of the radio-frequency front end, designing a real-time reconfigurable antenna capable of concentrating the radio-frequency energy along a specific spatial direction, requiring therefore less energy to be operated.
