Innatera brings Pulsar evaluation hardware to DigiKey
Illustrative image: Albert Stoynov on Unsplash
Innatera's Pulsar evaluation hardware is now available through DigiKey in the European Union, accompanied by a developer portal bringing together software, documentation and technical support. Availability in further countries is expected to follow before the end of the year. [1]
The setup with two boards pairs IgnisCore, carrying the processor, memory, USB programming interface and circuitry for monitoring power, with SensForge, which adds microphones, radar, motion and temperature sensors, Bluetooth Low Energy and expansion connections. It therefore provides real inputs for applications such as presence detection, audio recognition and machine monitoring, alongside a way to examine their energy consumption. [1]
Pulsar combines several kinds of processing on one chip: a fabric for spiking neural networks, a conventional accelerator for convolutional neural networks and a RISC-V CPU. That mix gives software a choice of where to run different parts of a sensing application, rather than requiring the whole workload to be expressed as a spiking network. The CPU handles control, with dedicated hardware available for signal processing and inference. [2]
In a spiking network, the timing of events carries information alongside which neurons respond, allowing a sound or changing sensor signal to be represented through activity distributed over time. The attraction for persistent sensing is that computation can follow activity in the input, although the energy used by a finished application also includes capturing and preparing that input. [2][3]
Through Talamo, developers can use a Python workflow with PyTorch integration, encoders that convert numerical inputs into spikes and decoders that turn network activity into an application result. The software also accommodates preprocessing, so a sensor pipeline can be developed and simulated before it is transferred to the processor. [3] Once Talamo has compiled the pipeline, Spine supplies the embedded toolchain and firmware needed to execute it. The portal collects those resources with examples and community support, alongside the hardware route for checking execution and processor power. [1]
