Technology

The Multi-Channel Multi-Band 500-1700nm spectral sensor chip

MantiSpectra has developed a unique proprietary technology¹ for reproducible, high-volume production of solid-state hyperspectral sensor chips, through the implementation of computational spectroscopy and the combined on-wafer fabrication of key components, including photodetectors and filters.

¹1WO2017118728A1, PCT/NL2020/050044, Dutch No. 2029194

Computational Spectroscopy

MantiSpectra proprietary technology captures the (bio)materials chemical fingerprints by making use of a series of pre-defined, non-linearly overlapping, frequency oversampled, multipeak channels. This unique and rich dataset combined with ML/AI algorithms enables accurate classification and quantification of (bio)materials characteristics.

For Illustrative Purposes: Each colour represent the frequency response of a single channel

Wafer integration

MantiSpectra technology makes use of InP wafers over which InGaAs photodetectors are built with integrated film filters. This monolithic integration at wafer level enables high signal-to-noise ratio, high responsivity, resistance to vibrations and high-volume manufacturing of low-cost and highly reproducible chips.

Applicability requirements

Material
Related

Works on liquids, solids and powders (not metal, not gases)

Parameter to be measured has to be organic¹, must have fingerprints in the covered optical range (e.g. 500-1700 nm) and concentration >~100 ppm

¹carbon-based substances, natural or biologically derived materials

Measurement
conditions
related

Optical sensing area (spot size of ~ 1cm2) assumed to be representative of the whole sample

Controlled Optical environment (e.g. minimal environment light)

Short measurement distance (< 5cm) => the longer the distance the more complex the optical system becomes

AI/ML model
Related

Models are material-dependent => models developed on a specific material cannot be re-used “as is” for others

Model accuracy is impacted by the reliability and representative distribution of the reference values

Model robustness depends on training²=> the more the models get trained considering samples and measurement conditions variability the more robust they become

²Spectroscopy is a secondary method: sensor measurements needs be correlated to independently obtained reference values