Internet-of-Things (IoT) devices. Since the storage density of battery technologies has not followed Moore's law scaling trends, IoT systems that rely entirely on power conversion from outside sources such as thermal, vibrational, light, or radio waves are needed. [8] High-performance edge computers, incorporating complex IoT functions such as multiple sensors, encryption/decryption, data storage to nonvolatile memory (NVM), and computing intensive algorithms, [9] draw several hundreds of microwatts of power using the latest IC technology nodes. [10] Edge computing systems with higher memory data rates would be needed for artificial intelligence (AI) applications. [11] Several hundred of milliwatts of power may be required to improve performance for neuromorphic computing. [12] Power converters that can provide highest power in the smallest possible footprint and can be integrated with multiple chips would enable greater functionality. As radio frequency (RF) power converters suffer from low efficiency for chip size <1 mm due to reduced antenna size (l ≪ λ), light energy as a power source can provide much higher efficiency and higher power density in a much smaller footprint than RF devices. In addition to generating high power density, these photovoltaic (PV) devices can provide a stable and sufficient output voltage for on-chip circuitry (>3 V) to drive the nonvolatile memory, charge an on-chip battery, and other devices, without using the less efficient charge pump circuit to ramp up the voltage. There is a great interest in using micro-PVs to power and also communicate with IoT chips. [6,13-16] However, there are two main challenges with small edge computers using photovoltaic energy harvesting and power conversion. 1) Integration: no manufacturable process has been demonstrated for integrating micro-PV and other chips into one system, and 2) power density: existing micro-PVs are unable to meet the above power density and efficiency requirements. We developed solutions to both of these problems in this work. In prior works, the integration process was done using nonmanufacturable processes, including chip stacking and wire bonding between chips. [7,17-22] A high-throughput