Efficient indoor light harvesters introduce a new design paradigm to Internet of Things (IoT) devices to maximize their ability to process, sense, and communicate data. Current employment of batteries or direct connection to the power grid can supply high power, but extended operation will require high maintenance and limit the deployment. By 2025, there may be as many as 40 billion devices connected to the internet and in most cases deployed indoors, all of which will require an energy source. We have developed a photovoltaic system to power a wireless sensor and to actually implement AI, such as a neural network utilising exclusively ambient light as power source. Wireless sensor nodes were powered by over 30% efficient dye-sensitized solar cells with an adapted sensitizer combination and with electrolyte redox mediator based on coordination copper complexes. Using a pretrained artificial neural network, we show that the collection in the magnitude of 10¹⁵ photons is needed for each inference. A 64 cm² photovoltaic area suffices to harvest enough energy to train and verify one epoch of MNIST datasets within less than 24 hours. Indoor light harvesters will lead to a new generation of self-powered
IoT, capable of advanced machine learning.¹

To enable future larger area production, the liquid electrolyte had to be replaced by a solid charge-transport material.Fast charge separation in a variety of colored organic dyes and tuneable energy levels in Cu(II/I) redox systems combined with negligible recombinative processes allow DSCs to maintain a high photovoltage, which is especially important under ambient light.² Further, co-sensitization of organic dyes enables absorption over a broad spectral range in the visible range and therefore adaption to the majority of light sources, which emit in the visible range. As a result, these DSCs outperform organic photovoltaics, silicon and thin-film GaAs technology under ambient light. Stable and
record-breaking solar cell efficiencies of over 11,5% under 1 sun were reached. In ambient conditions, 1000 lux indoor illumination the ”zombie” photovoltaics showed a power conversion efficiency over 34% and 103 μW/cm2 power output at 1000 lux.³

The conversion of ambient light offers broadly available energy and paves the way to extensive
implementation of self-powered devicesDSCs will be a key element in harvesting environmental energy. Given that autonomous and smart IoT devices are envisioned to be deployed widely in the ambient environment, combining high efficiency and low cost with non-toxic materials is of paramount importance to sustainability