This paper presents a new file system, BPFS, and a hardware architecture designed for byte-addressable, persistent memory (BPRAM), such as phase change memory (PCM). BPFS provides strong reliability guarantees and better performance than traditional file systems, even when both are run on BPRAM. The hardware architecture enforces atomicity and ordering guarantees required by BPFS while maintaining the performance benefits of L1 and L2 caches. BPFS uses a new technique called short-circuit shadow paging to provide atomic, fine-grained updates to persistent storage. This technique allows BPFS to perform in-place updates and avoid unnecessary copies, resulting in faster performance. BPFS is evaluated on DRAM and simulated on PCM, showing that it is significantly faster than traditional file systems like NTFS. The paper discusses the design principles of BPFS, including exposing BPRAM directly to the CPU, enforcing ordering and atomicity in hardware, and using short-circuit shadow paging for fast and consistent updates. The paper also discusses the limitations of BPFS, including the need for additional atomic write primitives and the challenges of supporting memory-mapped files. Finally, the paper discusses the hardware support required for BPRAM, including phase change memory, wear leveling, and write failures.This paper presents a new file system, BPFS, and a hardware architecture designed for byte-addressable, persistent memory (BPRAM), such as phase change memory (PCM). BPFS provides strong reliability guarantees and better performance than traditional file systems, even when both are run on BPRAM. The hardware architecture enforces atomicity and ordering guarantees required by BPFS while maintaining the performance benefits of L1 and L2 caches. BPFS uses a new technique called short-circuit shadow paging to provide atomic, fine-grained updates to persistent storage. This technique allows BPFS to perform in-place updates and avoid unnecessary copies, resulting in faster performance. BPFS is evaluated on DRAM and simulated on PCM, showing that it is significantly faster than traditional file systems like NTFS. The paper discusses the design principles of BPFS, including exposing BPRAM directly to the CPU, enforcing ordering and atomicity in hardware, and using short-circuit shadow paging for fast and consistent updates. The paper also discusses the limitations of BPFS, including the need for additional atomic write primitives and the challenges of supporting memory-mapped files. Finally, the paper discusses the hardware support required for BPRAM, including phase change memory, wear leveling, and write failures.