Conventional battery management systems (BMSs) for electric vehicles (EVs) are designed in an ad hoc way, causing the supply of EVs to fall behind the market demand. A well-designed and combined hardware-software architecture is essential for the efficient management of a large-scale battery pack that may consists of thousands of battery cells as in Tesla Motors and GM Chevy Volt. We propose a Dependable, Efficient, Scalable Architecture (DESA) that effectively monitors a large number of battery cells, efficiently controls, and reconfigures, if needed, their connection arrangement. DESA supports hierarchical, autonomous management of battery cells, where a global BMS orchestrates a group of local BMSs. A local controller on each local BMS autonomously manages an array of battery cells, and the global controller reconfigures the connectivity of such battery-cell arrays in coordination with the local controllers. Also, DESA allows individual arrays and local BMSs to be selectively powered-off for energy savings. The performance of this energy-saving capability is modeled and evaluated using a Markov chain. Our evaluation results show that DESA effectively tolerates battery-cell failures by an order-of-magnitude-while achieving 7.4 × service cost savings-better than a conventional BMS. This superior performance not only extends the battery life significantly, but also provides the flexibility in supporting diverse electric power demands from a growing number of on-board applications.