Memory and information processing in neuromorphic systems (1506.03264v1)

Abstract: A striking difference between brain-inspired neuromorphic processors and current von Neumann processors architectures is the way in which memory and processing is organized. As Information and Communication Technologies continue to address the need for increased computational power through the increase of cores within a digital processor, neuromorphic engineers and scientists can complement this need by building processor architectures where memory is distributed with the processing. In this paper we present a survey of brain-inspired processor architectures that support models of cortical networks and deep neural networks. These architectures range from serial clocked implementations of multi-neuron systems to massively parallel asynchronous ones and from purely digital systems to mixed analog/digital systems which implement more biological-like models of neurons and synapses together with a suite of adaptation and learning mechanisms analogous to the ones found in biological nervous systems. We describe the advantages of the different approaches being pursued and present the challenges that need to be addressed for building artificial neural processing systems that can display the richness of behaviors seen in biological systems.

• The paper presents a comprehensive survey of neuromorphic architectures that integrate memory and processing to achieve adaptive and efficient computing.
• It details diverse implementations—from serial clocked designs to asynchronous mixed-mode systems—highlighting impacts on scalability and energy efficiency.
• The research underscores practical implications for sensory processing and autonomous systems, paving the way for future AI and cognitive computing advances.

Memory and Information Processing in Neuromorphic Systems

The paper "Memory and Information Processing in Neuromorphic Systems" by Indiveri and Liu provides a comprehensive survey of brain-inspired processor architectures, examining various implementations ranging from serial, clocked multi-neuron systems to massively parallel asynchronous systems. These architectures are distinguished from traditional von Neumann processors by the co-location of memory and processing, a feature inspired by biological neural systems.

Overview of Neuromorphic Architectures

The paper presents an overview of neuromorphic architectures, categorizing them by serial clocked implementations and asynchronous systems. These include digital systems, mixed-mode analog/digital systems, and models that incorporate biological-like neurons and synapses with adaptation and learning mechanisms. Unlike von Neumann architectures with centralized memory, neuromorphic systems distribute memory functionality with processing units, aligning with neural network synapses that house both memory storage and computing operators.

Numerical Results and Claims

The researchers provide examples of existing systems such as SpiNNaker, TrueNorth, and NeuroGrid, each with different design choices affecting scalability, energy efficiency, and computational capabilities. TrueNorth, for instance, features 4096 cores and occupies 4.3 cm² in a 28nm CMOS process, illustrating the scale and design considerations necessary for implementing parallel, distributed systems.

Implications and Future Directions

Addressing memory-related constraints—a prevalent issue in traditional architectures—neuromorphic systems offer potential solutions replicating the fault-tolerant, adaptive computing found in biological systems. The integration of memory with processing could alleviate significant bottlenecks by enabling architectures that are intrinsically energy-efficient and robust.

The practical implications of this research are notable, with potential applications in sensory processing and autonomous systems. Neuromorphic systems' ability to model real-time interaction with dynamic environments marks them as significant for future developments in areas requiring low-power consumption and high adaptability.

Theoretically, these systems contribute to understanding computational paradigms beyond the classical scope, offering insights into integrating new technologies like resistive memories and memristors to expand functionality.

Conclusion

As the field advances, interdisciplinary collaboration among neuroscientists, material scientists, and engineers will be crucial for the continued development of neuromorphic systems. These systems represent a promising avenue toward achieving computing paradigms that emulate the efficiency and adaptability of the human brain, with prospective breakthroughs in artificial intelligence and cognitive computing.

In summary, this paper delineates the current landscape and future potential of neuromorphic systems, pioneering a shift towards architectures that transcend traditional computing limitations through biologically inspired design principles.