Carbon nanotube networks (CNNs) have become suitable for applications as sensors, interconnects and flexible electronics through recent development of large-scale fabrication, control of nanotube density, chirality and alignment [1-2]. However, due to the complexity of their structure, performance limits and reliability of CNN-based devices are not well understood. In this study, we combine comprehensive experiments and computational modeling to analyze high-field transport, thermal dissipation, and breakdown in CNN devices. We fabricated thin film transistors with various lengths (5-50 μm), widths (5-150 μm) and nanotube densities, on SiO2/Si substrates with Cr/Au electrodes. Importantly, we examined all-metallic, all-semiconducting and mixed-chirality CNNs obtained by density gradient ultracentrifugation [2]. Devices were electrically characterized, then imaged by AFM, SEM and infrared (IR) thermal microscopy up to electrical breakdown. The computational model randomly generates tubes with a predefined distribution of orientation, position, density and type based on the experiments. Current and power dissipation and temperature profile were calculated by solving electrical transport equations at each network element [3], self-consistently with the thermal transport equations. CNN breakdown in simulations occurs when all (...) tubes are burned out and removed from the network, to mimic the experimental breakdown at high voltage [4]. We find that average breakdown voltages scale linearly with device length with a slope of 1.6 V/μm for mixed-networks. However, we found no strong influence of the gate voltage on the CNN breakdown voltage, even for all-semiconducting networks. Simulations offer the insight that CNN breakdown is initiated at nanotube junctions and that a zigzag burning path is formed along the width of the network, consistent with our SEM imaging [4], but too fast to be imaged in real-time by the IR camera. In addition, average breakdown voltage and its standard deviation decrease (unlike maximum power capacity that increases) when nanotube density or network purity increase (metallic or semiconducting tube percentage becomes very high). For 10x5 μm devices, for example, the breakdown voltage reduces from 90 to 45 V as density increases from 2 to 10 tubes/μm2. Furthermore, networks that are very sparse or very pure tend to break instantly, instead of degrading gradually. The results of this study offer a comprehensive look at the complex reliability phenomena in CNN devices for practical applications. 1. X. Ho, et al., Nano Lett. 10, 499 (2010). 2. M. Engel, et al., ACS Nano 2, 2445 (2008). 3. A. Behnam, et al., Phys. Rev. B 75, 125432 (2007). 4. D. Estrada and E. Pop, Appl. Phys. Lett. 98, 073102 (2011).