Neuromorphic Architecture
Neuromorphic computing represents the third generation of neural networks, is biologically plausible and is fundamentally different from conventional neural networks and related AI accelerators. Compared to conventional Artificial Neural Networks (ANNs), they can provide higher computational power and can learn and compute in ways closely resembling a biological brain.
Cyceera has invented (patents pending) and is developing a novel modular and scalable Neuromorphic device architecture that performs both continuous real time autonomous machine learning and inference on the same device. The patent cover a range of novel features. Inherent architectural features facilitate the implementation of Explainable Artificial Intelligence (XAI) allowing for interpretability, transparency, accountability and trustworthiness; key requirements for next generation human-machine collaboration systems and applications.
Cyceera’s Explainable Neuromorphic technology is derived from our research and development of our highly parallel dataflow processor and cognitive processing technology, for which patents have been granted internationally. It will provide a step change in performance and capabilities over second generation AI devices and current Neuromorphic “black boxes” and hence un-interpretable offerings. In addition, its unique small, low power, highly parallel, fault tolerant scalable architecture forms the common denominator in all next generation AI systems and therefore targets a wide range of new and emerging applications, such as wearable SMART sensors and Explainable AI applications.
Cyceera’s Machine Learning technology provides a completely new computing platform, designed from the ground up for machine intelligence. By creating custom IP / hardware for Machine Learning, Cyceera is able to provide the means for its customers to tackle new research and increase their potential to do so much more with Machine Learning powered applications. Cyceera’s Machine Learning technology can deliver orders of magnitude better-optimized performance per watt for Machine Learning than current solutions.
KEY FEATURES
Brain inspired single device that is not cloud dependent.
Low complexity, modular and scaleable digital architecture based on Sparse Distributed Representations (SDR).
Highly parallel and hierarchical architecture provide high speed neuron & synapse updates.
Low power energy efficient architecture with low computational overheads.
Real time unsupervised learning & incremental learning without the need for large datasets. Implements both spatial and temporal learning and sequence generation.
Handles fuzzy data & concepts.
BENEFITS
Small silicon area that doesn’t require additional processors, software and external memory – reduced Bill Of Materials (BOM) and hence system costs;
Orders of magnitude increase in performance through autonomous real time feature extraction and learning – reduced running costs and quicker time to market;
Scalable solution that provides upgrade and software migration paths for easier product development and introduction. This in turn reduces product development and commercialisation risks as development is based on known and tested technology and so reduces production costs and time to market.