Neuromorphic Intelligence

Ultimate AI Solutions

Inspired by the working and cognitive mechanism of neurons, neuromorphic intelligence uses software/hardware co-design and computational modelling to achieve extreme power efficiency. Neuromorphic devices can be more intelligent than humans and is the most possible way leading to general intelligence.

Insects can track, navigate, and avoid objects. They are capable of tasks such as inference, decision-making. In contrast, the human brain is far more complex and powerful.

As the basis of neuromorphic intelligence, neuromorphic chips imitate the functions of the brain, i.e., information processing, transmitting, and learning. With ultra-low power consumed, neuromorphic chips complete the tasks of sensing, learning, memorizing, decision-making, etc. making intelligence smarter.

Roadblocks lie ahead for AI

Hardware Bottlenecks

  • Transistors are projected to reach their limits. The Moore’s law is coming to an end.
  • In the face of tons of data, the Von Neumann architecture wastes cost and power on data transmission.
  • The computational power is limited at the edge. The scenarios that require high computational power rely on cloud computing, The process of transmitting data between cloud and the edge wastes enormous power and cost.

Algorithm Bottlenecks

  • Vast trove of data needs additional memory resources.
  • A large-scale system requires computing resources and power.
  • Applications of weak AI are restrained and are only available for limited tasks.

Features of Neuromorphic Chips

New Computing Mechanism

Event-driven computing that is based on sparse communication

New Architecture

Novel synchronize computing, distributed kernel/memory

Cutting Edge Algorithm

Spatial temporal computing using spiking neural network

Neuromorphic computing,
New Paradigm in Post-Moore era

Breakthrough in architecture and algorithm

Traditional
Architecture
Neuromorphic
Architecture
Weak AIGeneral AI
von Neumann ArchitectureSynchronous Parallel Processing
Single Sensor Multimodality Sensor
Traditional
Architecture
Weak AI
von Neumann Architecture
Single Sensor
Neuromorphic
Architecture
General AI
Synchronous Parallel Processing
Multimodality Sensor
Data intelligence driven by model learningCognitive bio-inspired neuromorphic intelligence
Massive data, High quality labelingFew-shot learning, labeling
Low adaptive ability, highly dependency on model.Unsupervised learning, adaptive ability
High computational costs, Power hungryLow computational resources costs, Low Power
Weak dependency on temporal sequence Strong temporal sequence dependency, general solution to classical application scenario.
Data intelligence driven by model learning
Massive data, High quality labeling
Low adaptive ability, highly dependency on model.
High computational costs, Power hungry
Weak dependency on temporal sequence
Cognitive bio-inspired neuromorphic intelligence
Few-shot learning, labeling
Unsupervised learning, adaptive ability
Low computational resources costs, Low Power
Strong temporal sequence dependency, general solution to classical application scenario.

Advantages

Event-driven

Power consumption reduced by 100-1000 times

Asynchronous

Real-time increased by 10-100 times

High temporal sequence dependency

Dynamic information processing

General Artificial Intelligence

Lower cost up to 100 times

Continuous Innovation with cloud-edge fusion solution

2021–2022

Sensor Node I

Low computational costs on the edge

AI computation nodes

Smart Home
Smart Toy
CHIPS – SPECK, XYLO,
2022–2023

Sensor Node II

High computational costs on the edge

Smart Security
CHIPS - DYNAP-CNN,
2023

Sensor Fusion

Multi-sensory fusion computing

Autonomous Driving

High-speed autonomous obstacle avoidance

Vehicle-road coordination

Drones
CHIPS - DVS-SLAM,
2023–2024

Edge Cloud

Analog computation, neuromorphic near-memory computing

AR/VR

Machine Perception Optical Flow Localization Visual Navigation Control

Robots
CHIPS - MULTI-CORE DYNAP, DYNAP-M, XYLO-M,
Close
Back To