Integrating convolutional neural networks accelerators in commercial MCUs
keywords ARTIFICIAL INTELLIGENCE, CONVOLUTIONAL NEURAL NETWORKS, DEEP LEARNING, DEEP NEURAL NETWORKS, EMBEDDED SYSTEMS, ENERGY EFFICIENCY, FIRMWARE DEVELOPMENT, LOW POWER, MICROCONTROLLERS, SOFTWARE
Reference persons DANIELE JAHIER PAGLIARI
External reference persons Alessio Burrello (University of Bologna)
Research Groups DAUIN - GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA, ELECTRONIC DESIGN AUTOMATION - EDA, GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA
Thesis type EXPERIMENTAL, SOFTWARE DEVELOPMENT
Description The use of dedicated hardware accelerators for the execution of deep neural network inference leads to > 100x efficiency improvement (think of Google's TPUs or Cerebras). However, using accelerators effectively requires the development of flexible yet efficient software support libraries, able to fully exploit the power of the custom hardware while hiding the details of complex interfacing and data transfer issues to the user. Moreover, depending on the characteristics of the neural network, using only the dedicated hardware might not be optimal, and even better efficiency could be obtained by combining CPU(s) and accelerator execution.
The purpose of this thesis is to enable the use of a new accelerator for neural networks called RBE (Reconfigurable Binary Engine), within the commercial low-power microcontroller GAP8 by GreenWaves Technologies. In particular, the candidate will: i) study the architecture of the RBE; ii) define the integration requirements; iii) write a complete software library enabling the use of RBE within GAP8; iv) develop a tool able to automatically transform a neural network model into executable code that combines the GAP8 CPUs and the accelerator to achieve the maximal efficiency. The resulting system will be evaluated and compared against various known architectures from the state of the rt.
Required skills Required skills include C and Python programming. Furthermore, a basic knowledge of computer architectures and embedded systems is necessary. Desired (but not required) skills include some familiarity with basic machine/deep learning concepts and the corresponding models.
Notes Thesis in collaboration with Prof. Luca Beniniís research group at the University of Bologna and ETH Zurich. The thesis can be carried out either in Torino or in one of the other two universities.
Deadline 13/12/2022 PROPONI LA TUA CANDIDATURA