Dirk Kutscher

Our paper on AdaptQNet: Optimizing Quantized DNN on Microcontrollers via Adaptive Heterogeneous Processing Unit Utilization has been accepted at ACM MobiCom-2025.

Abstract

There is a growing trend in deploying DNNs on tiny micro-controller (MCUs) to provide inference capabilities in the IoT. While prior research has explored many lightweight techniques to compress DNN models, achieving overall efficiency in model inference requires not only model optimization but also careful system resource utilization for execution. Existing studies primarily leverage arithmetic logic units (ALUs) for integer-only computations on a single CPU core. Floating-point units (FPU) and multi-core capabilities available in many existing MCUs remain underutilized.

To fill this gap, we propose AdaptQNet, a novel MCU neural network system that can determine the optimal precision assignment for different layers of a DNN model. AdaptQNet models the latency of various operators in DNN models across different precisions on heterogeneous processing units. This facilitates the discovery of models that utilize FPU and multi-core capabilities to enhance capacity while adhering to stringent memory constraints. Our implementation and experiments demonstrate that AdaptQNet enables the deployment of models with better accuracy-efficiency trade-off on MCUs.

References

Yansong Sun, Jialuo He, Dirk Kutscher, Huangxun CHEN; AdaptQNet: Optimizing Quantized DNN on Microcontrollers via Adaptive Heterogeneous Processing Unit Utilization; The 31st Annual International Conference On Mobile Computing And Networking (MobiCom 2025)

One of the papers recenty presented at ICN-2014 described an interesting IoT implementation and corresponding experiment with CCN-Lite on the RIOT platform.

Previously, ICN has been perceived as providing conceptual benefits such as

• simplified, natural APIs to developers;
• increased robustness through caching;
• facilitating data fusion through hop-by-hop replication;
• reduced network stack layering; and
• inherent auto-configuration.

The authors describe their implementation of CCN-Lite on RIOT and their approach to realize IoT communication in a 60 node testbed. The idea is to apply Reactive Optimistic Name-based Routing (RONR), i.e., an ICN name-based forwarding approach to send requests for named information in an IoT network using a hybrid flooding/unicast approach.

Some results of their comparison:

• 70% less ROM, 80% less RAM usage by the stack implementation (compared to a RPL/6lowpan implementation);
• 50% reduction of transmitted packets thanks to RONR and ICN caching.

Some pointers for further reading:

•  Emmanuel Baccelli, Christian Mehlis, Oliver Hahm, Thomas C. Schmidt, Matthias Wählisch; Information Centric Networking in the IoT: Experiments with NDN in the Wild; ACM ICN-2014; September 2014; [paper], [presentation]
• CCN-Lite
• RIOT -- the friendly OS for the IoT
• ACM SIGCOMM ICN-2014