memoryhard

Memoryhard is a term used in computer science to describe an algorithm or function whose performance is constrained primarily by memory usage rather than by CPU speed alone. In practice, memoryhardness is achieved by requiring a large, often randomly accessed memory footprint during computation. This design makes it difficult to accelerate computation with specialized hardware that has relatively high compute power but limited memory per core, and it is widely used to strengthen password hashing and key derivation.

Memory-hard functions (MHFs) are intended to resist efficient time-memory trade-offs. They typically require allocating substantial memory for each computation and use memory access patterns that hinder simple recomputation. Because memory bandwidth and capacity dominate the cost, attackers using GPUs or ASICs face diminishing returns when trying to optimize brute-force or precomputation attacks.

Common examples include scrypt and Argon2. Argon2, the winner of the Password Hashing Competition, provides tunable parameters for memory, time, and parallelism, and offers variants such as Argon2d, Argon2i, and Argon2id. These MHFs are widely used for password hashing and as key derivation functions, with the goal of making mass cracking more expensive while keeping legitimate verification feasible on ordinary hardware.

Advantages of memory-hard designs include increased resistance to parallelized cracking and better scalability with hardware that features high memory bandwidth. Limitations include higher memory requirements, potential performance variability due to memory access patterns, and the need for careful parameter selection to balance security and practicality.