Spiking Neural Network Chip for Smarter Sensors

By mimicking the way the brain operates, neuromorphic processors can expend dramatically less energy than conventional technology for certain applications. Now, Dutch firm Innatera has launched what it calls the world’s first commercially available neuromorphic microcontroller, in the hope of spurring mass-market adoption of this emerging technology.

Innatera says its new chip, Pulsar, can deliver as much as 100 times lower latency and 500 times lower power consumption that conventional processors used for artificial intelligence applications. “Most AI accelerators today have to deal with a tradeoff between performance and power,” says Sumeet Kumar, co-founder and CEO of Innatera. “They either run simplified AI models to consume less power, or ramp up their accuracy and the amount of power they need. With Pulsar, you don’t have to give up anything.”

Neuromorphic Chips Mimic Brain Function

Neuromorphic devices often imitate the workings of the brain in a variety of ways. For instance, whereas conventional microchips use clock signals fired at regular intervals to coordinate the actions of circuits, neuromorphic architecture often “spikes”—that is, generates an output signal—only after it receives a certain amount of input signals over a given time.

A key application often envisioned for neuromorphic technology is to implement similarly brain-inspired neural networks, the main AI systems in use today. In addition, spiking neuromorphic devices only rarely fire spikes, so they shuffle around much less data than the electronics that typically run neural networks. As such, neuromorphic hardware in principle requires much less power and communication bandwidth for artificial intelligence applications.

So far, neuromorphic devices have not found widespread use. Now Innatera hopes that Pulsar, launched on 21 May, can overcome barriers that neuromorphic computing has long faced to commercialization.

The Pulsar chip possesses a hybrid analog-digital architecture. In addition to 12 digital cores for spiking neural networks, it also has four analog ones, with silicon circuits making up the spiking neurons and interconnecting synapses of each core.

“The analog spiking fabric offers extremely high energy efficiency, while the digital spiking fabric offers more programmability and configurability while still offering very good energy efficiency,” Kumar says. Developers can pick which set of cores they want to load their models onto depending on their needs, he explains.

Each Pulsar chip also has an accelerator for convolutional neural networks (which are often used for image recognition and natural language processing) that supports 32 multiply-accumulate (MAC) operations. In addition, each chip possesses a fast Fourier transform accelerator for efficient low-power signal processing. Each Pulsar also incorporates a 32-bit RISC-V CPU that can run at up to 160 MHz for systems management, as well as a range of standard sensor interfaces and other components. “All of this is integrated into a tiny chip of 2.8 by 2.6 millimeters,” Kumar says.

What Makes Pulsar Unique in AI Sensors?

What sets Pulsar apart from other neuromorphic devices, such as BrainChip’s Akida Pico, “is not just building a neuromorphic core, but also the rest of the system around it,” Kumar says. “In the industry, there’s a lot of emphasis on inference, but when their neuromorphic cores speak with the rest of their systems, you see them burning power moving data in and out, and all the energy gains they can bring to the table quickly become irrelevant. We built Pulsar as an engine for efficient processing, not just efficient inference.”

By integrating all these functions together, “it’s the only chip a sensor needs to process data,” Kumar says. This can simplify overall device design, which can reduce the need for complex data signal processing pipelines, speed up development and time to market, lower maintenance costs, extend battery life and enable sub-millisecond analysis times.

With sub-milliwatt power consumption, “Pulsar enables always-on processing of sensor data, even in devices radically constrained by power,” Kumar says. For example, it can enable radar-based presence detection with as little as 600 microwatts and audio scene classification with just 400 microwatts. In comparison, similar applications using conventional electronics consume 10 to 100 milliwatts, he notes.

Pulsar is designed for ultra-low-power AI sensor applications in consumer, industrial and IoT settings. For example, it may find use in smart doorbells, which currently detect people by using cameras or infrared sensors to pick up motion. “This makes them susceptible to be triggered by a flag fluttering in front or a pair of headlights going down the street, draining their batteries” Kumar says. “Most smart doorbells are advertised as having a battery life of three months, when realistically they need to be recharged every two to three weeks.”

Innatera is partnering with Japanese system-on-a-chip company Socionext to develop a radar-based sensor that can accurately detect people even if they are standing perfectly still, based on their body motions as they breathe. “It can ignore things like bushes moving in the wind,” Kumar says. “It can extend smart doorbell operations to 18 months per recharge. And since it’s not camera-based and doesn’t store data in the cloud, it helps protect privacy.”

A key obstacle that neuromorphic computing faces is the steep learning curve that confronts developers as they seek to run their models on these devices. As such, Innatera has released its Talamo software development kit to reduce this barrier to entry, with which developers can build spiking models from scratch in a PyTorch-based environment. “You should not need a neuromorphics PhD to run a neuromorphics solution on chips like these,” Kumar says.

The company is also launching a developer program, now open to early adopters, to provide hardware and software kits to a growing community of researchers. “The hope is to grow an ecosystem of neuromorphics applications, and to discover things so far not even thought about,” Kumar says.