Intern - France

6 months Internship – Evaluation of Spike Neural Network on Time-Series Sensor Data

Overview/Job Summary

Tra ditional AI hardware architectures (GPUs, standard microcontrollers) rely on von Neumann models where data must constantly shuttle between memory and processing units. For real-time processing of continuous time-series sensor data (e.g., IMU data, industrial vibration monitoring, biosignals, bio-inspired event streams), this introduces severe power, latency, and throughput bottlenecks.

Traditional AI hardware architectures (GPUs, standard microcontrollers) rely on von Neumann models where data must constantly shuttle between memory and processing units. For real-time processing of continuous time-series sensor data (e.g., IMU data, industrial vibration monitoring, biosignals, bio-inspired event streams), this introduces severe power, latency, and throughput bottlenecks.

Evaluation of Neuromorphic Edge Processors for Time-Series Sensor Analytics.

Neuromorphic architecture resolves these limitations by drawing direct inspiration from biological neural systems:

- Event-Driven Computing: Instead of processing dense frames or fixed time-window matrices at rigid clock intervals, neuromorphic chips process sparse, asynchronous temporal events (spikes) only when state changes occur.
- Extreme Energy Efficiency: Power is consumed almost exclusively when processing incoming events, enabling ultra-low-power edge deployment (sub-milliwatt to single-digit milliwatts).
- Ultra-Low Latency: Asynchronous event processing yields near-instantaneous inference delays, making it ideal for closed-loop edge control and instantaneous anomaly detection.
- In-Memory Computing & Temporal Dynamics: Co-locating memory and processing (via Spiking Neural Networks / SNNs) allows time-series dynamics to be captured directly through temporal integration and recurrent spiking mechanics.

This internship targets the benchmarking, performance characterization, and practical deployment of state-of-the-art neuromorphic platforms for real-world time-series sensor processing.

Internship Description

As part of the Software Engineering Team in Grenoble you will benchmark emerging neuromorphic processors against baseline microcontrollers/edge accelerators on realistic time-series sensor datasets. You mission is to:

- State-of-the-art study on SNN and input time-series data representations.
- Convert continuous or frame-based time-series sensor data into spiking representations (e.g., Delta modulation, rate encoding, direct temporal encoding). Design, train, and optimize Spiking Neural Networks (SNNs) for tasks such as temporal pattern recognition, state estimation, or dynamic anomaly detection using frameworks like Jax, snntorch, spikingjelly, or Lava.
- Compare neuromorphic edge performance against standard low-power edge baselines (e.g., Quantized DNN).
- Explore target neuromorphic hardware frameworks (e.g., SynSense Speck/Xylo, BrainChip Akida, Innatera, or Intel Loihi / Lava framework).
- Deploy trained SNN models onto physical neuromorphic hardware / evaluation boards. Quantify system-level metrics: dynamic power consumption, compute latency, energy-per-inference (J/inference), memory footprint, and task accuracy.

Qualifications

- Bac+5 program
- Strong proficiency in Python and deep learning frameworks (Jax, Tensorflow, PyTorch).
- Direct experience with time-series modeling, sensor data algorithms, or signal processing.
- Hands-on experience building structured AI agent workflows (state machines, tool use, dynamic routing).
- Solid working knowledge of SQL, time-series databases, and metric generation/KPI tracking.
- Familiarity with dynamic model evaluation, hyperparameter tuning, or AutoML loops.V
- Very good English and communication skills

Duration: 6 months

American R&D Center (ARDC) is a central R&D organization of the TDK Group. Its mission is to contribute into the design of new products within TDK with the ambition to create new product lines. ARDC is working closely with InvenSense Inc., a TDK Group Company, who is the world’s leading provider of MEMS sensor platforms. InvenSense’s vision of Sensing Everything™ targets the consumer electronics and industrial markets with integrated Motion, Sound, Pressure, Gas solutions.

InvenSense’s motion tracking, audio, and services can be found in many of the world’s largest and most iconic brands including smartphones, tablets, wearables, drones, gaming devices, internet of things, automotive products, and remote controls for smart TVs.

Both ARDC & InvenSense are headquartered in San Jose, CA, and have offices in Boston, China, Taiwan, Korea, Japan, France, Slovakia, and Italy.