(Source: spectrum)By mimicking the operation of the brain, neuromorphic processors can significantly reduce energy consumption in certain application scenarios compared to traditional technologies. Now, the Dutch company Innatera has launched what it claims to be the world’s first commercially available neuromorphic microcontroller, aimed at promoting the large-scale market application of this emerging technology.Innatera states that its new chip, Pulsar, can reduce latency to one-hundredth of that of traditional processors and consumes only one-five-hundredth of the power in artificial intelligence applications. Sumeet Kumar, co-founder and CEO of Innatera, stated: “Currently, most AI accelerators face a trade-off between performance and power consumption, either running simplified AI models to reduce power or increasing accuracy but at the cost of higher energy consumption. Pulsar requires no compromises.”Neuromorphic Chips Mimicking Brain FunctionsNeuromorphic devices imitate the brain’s working methods in several ways. For example, traditional microchips use a fixed rhythm clock signal to coordinate circuit actions, while neuromorphic architectures often operate through “pulses,” meaning they only produce output after receiving sufficient input signals over a certain period.One of the key applications of neuromorphic technology is to implement brain-inspired neural networks, which are the mainstream AI systems today. Additionally, pulse-based neuromorphic devices emit pulses at a very low frequency, resulting in far less data transmission than electronic systems running traditional neural networks. Therefore, theoretically, neuromorphic hardware has significantly lower power and communication bandwidth requirements in AI applications.Currently, neuromorphic devices have not been widely adopted. Innatera hopes to break the long-standing barriers to the commercialization of neuromorphic computing with the release of the Pulsar chip on May 21.The Pulsar chip features a hybrid analog-digital architecture, containing 12 digital cores for pulse neural networks and 4 analog cores, with each core’s pulse neurons and connecting synapses made from silicon circuits.Kumar stated: “The analog pulse structure is highly energy-efficient, while the digital pulse structure offers more programmability and configurability while maintaining good energy efficiency.” Developers can choose the core combinations to load models based on their needs.Moreover, the Pulsar chip includes a convolutional neural network accelerator supporting 32 multiply-accumulate (MAC) operations (commonly used for image recognition and natural language processing) and is equipped with a fast Fourier transform (FFT) accelerator for efficient low-power signal processing. Each chip also integrates a 32-bit RISC-V CPU with a maximum operating frequency of 160MHz for system management, along with a range of standard sensor interfaces and other components. Kumar stated: “All of this is integrated into a chip that measures just 2.8mm x 2.6mm.”The Uniqueness of Pulsar in AI SensorsKumar pointed out that the key difference between Pulsar and other neuromorphic devices like BrainChip’s Akida Pico is: “We have built not just a neuromorphic core, but a complete system around that core.” He added: “Currently, the industry mainly focuses on inference, but when the neuromorphic core interacts with other systems for data transfer, it consumes a lot of energy due to data handling, offsetting its original energy-saving advantages. Our goal in building Pulsar is to achieve overall efficient processing, not just efficient inference.”Kumar noted: “By integrating these functions, it becomes the only chip needed for sensor data processing.” This will simplify device design, reduce complex signal processing workflows, accelerate development and time-to-market, lower maintenance costs, extend battery life, and enable sub-millisecond data analysis.Due to its sub-milliwatt power consumption, “Pulsar can achieve continuous sensor data processing, even in devices with extremely limited power,” Kumar stated. For example, it can achieve radar-based presence detection with only 600 microwatts of power or audio scene classification with 400 microwatts. In contrast, systems using traditional electronic technologies to achieve similar functions typically require 10 to 100 milliwatts of power.Pulsar is designed for ultra-low-power AI sensor applications, suitable for consumer electronics, industrial, and IoT scenarios. For instance, it can be used in smart doorbells, where most smart doorbells detect motion through cameras or infrared sensors, but this is often falsely triggered by fluttering flags or car lights on the street, leading to rapid battery drain. “Although these devices claim battery life of up to three months, they typically need to be recharged every two to three weeks,” Kumar pointed out.Innatera is collaborating with Japanese SoC manufacturer Socionext to develop a radar-based sensor that can accurately detect micro-movements caused by breathing, even when a person is stationary. “It can ignore interference from things like bushes swaying in the wind,” Kumar stated, “and can extend the smart doorbell’s battery life to 18 months. Moreover, since it does not use cameras or upload data to the cloud, it offers better privacy protection.”A major obstacle facing neuromorphic computing is the steep learning curve developers must overcome when running models on such devices. To address this, Innatera has launched the Talamo software development kit to lower the entry barrier, allowing developers to build pulse neural network models from scratch in a PyTorch-based environment. Kumar stated: “Developing neuromorphic applications should not require a PhD in neuromorphology.”Additionally, the company has launched a developer program for early adopters, providing hardware and software toolkits to support the growing research community. “Our hope is to build an ecosystem of neuromorphic applications and discover some currently unimagined innovative applications,” Kumar stated.