公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容来自 spectrum 。 通过模拟大脑的运行方式，神经形态处理器在某些应用场景下比传统技术能显著降低能耗。如今，荷 兰公司Innatera推出了号称全球首款商用的神经形态微控制器，旨在推动这一新兴技术的大规模市场 应用。 Innatera表示，其新芯片Pulsar可将延迟降低至传统处理器的百分之一，并在人工智能应用中仅消耗 其五百分之一的功耗。Innatera联合创始人兼CEO Sumeet Kumar表示："目前大多数AI加速器都面 临性能与功耗之间的权衡，要么运行简化的AI模型以降低功耗，要么提高精度但增加能耗。而Pulsar 不需要任何妥协。" 神经形态芯片模拟大脑功能 神经形态设备在多个方面模仿大脑的工作方式。例如，传统微芯片使用固定节奏的时钟信号来协调电 路动作，而神经形态架构则常通过"脉冲"来工作，即在一定时间内接收到足够输入信号后才会产生输 出。 神经形态技术的关键应用之一，是用于实现受大脑启发的神经网络，也就是如今主流的AI系统。此 外，脉冲式神经形态设备发射脉冲的频率很低，因此传输的数据量远少于运行传统神经网络的电子系 统。因此，理 ...

Core Viewpoint - Innatera has launched the world's first mass-market neuromorphic microcontroller, Pulsar, designed for sensor applications, which significantly reduces latency and power consumption compared to traditional AI processors [2][3]. Group 1: Product Features - The Pulsar chip features a heterogeneous architecture that combines analog and digital neuromorphic modules with traditional convolutional neural network (CNN) accelerators and RISC-V cores [2]. - It achieves a latency reduction of 100 times and a power consumption reduction of 500 times, with a chip size of 2.6 x 2.8 mm and a manufacturing cost of less than $5 [2]. - The analog neural network (ANN) core processes time-series data efficiently without complex models, operating with a latency of just 1ms and power consumption below 1mW [3]. Group 2: Market Context - The sensor shipment volume reached 38 billion units last year and is projected to grow to 60 billion units by 2030, necessitating edge processing due to the speed of data generation [2]. - Current microcontroller models are limited, requiring developers to balance functionality, accuracy, and power consumption [2]. Group 3: Performance Metrics - For wireless earbuds, the inference power consumption for audio classification has been reduced by 100 times to 400 µW while maintaining over 90% accuracy, and the model size has shrunk by 33 times [4]. - In voice recognition, the inference power consumption has decreased by 88 times, while radar-based gesture recognition shows a 42 times reduction in power consumption compared to CNN accelerators [4]. Group 4: Development Tools - The Talamo SDK is designed to interact with PyTorch, facilitating a familiar environment for developers and simplifying the mapping of models to the chip architecture [5]. - Innatera plans to launch a developer program and a neuromorphic development board, aiming to create a collaborative ecosystem for neuromorphic AI [5].