University of MaineElectrical and Computer Engineering

NEST a vector memory that rewires itself around the queries you actually make

ACM/IEEE International Symposium on Low Power Electronics and DesignISLPED 2026CORE RANK A
LARGE-FORMAT COMPOSITE PRINTER THERMAL VIEW THERMAL CAMERA ON-BOARD EDGE DEVICE stores bead signatures, runs every lookup Each new bead is checked against stored references. Thousands of lookups per print.
Fixed index standard
the map never changes
–hops
TIME
ENERGY
NEST learns
rewires after every search
–hops
TIME
ENERGY
Measured on an NVIDIA Jetson edge computer, five benchmark datasets
14x to 46x
faster search than Qdrant, Weaviate, Chroma and Milvus on repeated queries
up to 99%
less energy per search, with accuracy on par with the other databases
0 rebuilds
new data streams in one item at a time, and the index keeps adapting
Numbers are from the ISLPED 2026 paper (NYTIMES dataset, prioritized query workload, NVIDIA Jetson AGX Orin). The printing scenario above is an illustration of the intended use, not an experiment from the paper.
Funded by the National Science Foundation
For more details, contact
Sarah Glatter, lead student researcher sarah.glatter@maine.edu·Prabuddha Chakraborty, SIEGE Lab PI prabuddha@maine.edu