Nowadays, efficient and remote health monitoring is becoming increasingly important, given both the ageing of population and the combined action of an increase in obesity level and cardiovascular diseases. The healthcare industry is becoming more reliant on new methods to monitor and treat patients. This, along with an increased interest in fitness and wellness, is calling for more affordable, precise and wearable health monitoring devices. In this context, photoplethysmography (PPG) appears to be a key technology allowing non-invasive monitoring of vital biological indicators such as the heart rate, the blood oxygen saturation, the respiration rate and the arterial blood pressure. A standard PPG system comprises pulsed LEDs synchronized with a photosensor and a processing chain. The LEDs diffuse light in the human skin. Processing the signal held by the diffused light allows the extraction of the vital parameters. Despite the great potential behind the PPG technology, the fairly large power burnt by the LEDs still represents a serious challenge towards truly continuous PPG operations, limiting its practical exploitation. State-of-the-art PPG sensors, both in academia and in commercial products, still follow a quite standard design paradigm. Indeed, they rely on off-chip photodiodes and relatively standard circuitry. The commercially available smartwatches and wearables fall short of meeting customer requirements in terms of reliability, precision and battery lifetime. In this regard, we should not expect any dramatic improvement unless there are fundamental changes in the PPG sensor technology. This is particularly true on the photosensor side, since its parasitic capacitance represents one of the limiting factor in terms of power/noise. Pinned-photodiode (PPD) are today the key ingredients of CMOS image sensors, thanks to the low dark current, low noise and large sensitivity operations. Several markets including security, scientific and medical imaging are relying today on this technology. The excellent performance of a PPD makes it particularly interesting for the PPG application. Indeed, the LED power can be reduced provided the noise floor is decreased proportionally. In this work a truly micropower PPG sensor combining an array of double transfer gates (TG) PPDs together with an ultra-low noise read-out chain is presented. Compared to conventional solution, this work achieves the same signal-to-noise ratio (SNR) at a significantly lower LED power. The PPDs array is implemented on the same chip together with the analog front-end including the analog-to-digital conversion. The full CMOS integration allows a dramatic reduction of the parasitic capacitance at the sense node leading to a larger conversion gain and a lower noise. This approach also provides higher miniaturization and lower cost compared to traditional solutions with off-chip PDs. The use of an array additionally enables spatial averaging leading to further noise reduction. Consequently, the LED power can be reduced dramatically. The chip, implemented in a 0.18 µm CMOS Image Sensor (CIS) process, features a total input referred noise of 0.68 e-rms per PPD, independently of the input light, and achieves a 4.6 µW total power consumption, including 1.97 µW LED power, at 1.38 bpm heart rate average error. Compared to the most recent state-of-the-art works this means more than an order of magnitude in power reduction.