128bitisation

128bitisation is a term used to describe the trend toward widening core computational elements to 128-bit width, including data paths, registers, and addressing. It is not an official standard, but a descriptive label for the gradual shift in hardware and software design toward wider 128-bit operations to handle larger numbers, more precise vectors, and expanded address spaces.

In practice, 128-bitisation covers several layers: 128-bit integer and floating-point data types and arithmetic in programming languages; 128-bit general-purpose registers or the ability to operate on 128-bit vectors; memory interfaces and pointers that can reference a 128-bit address space; and compiler and toolchain support to generate and optimize 128-bit code. It often proceeds alongside advances in SIMD units, where 128-bit lanes are common, and with growing use of 128-bit keys or blocks in cryptography.

Current status: There are already widespread 128-bit vector units (for example 128-bit SIMD registers in various architectures) and languages provide 128-bit integers as builtins. A full transition of general-purpose data paths to 128-bit width remains speculative and architecture-specific, with significant design and cost considerations. Some researchers and vendors discuss 128-bit addressing in future CPUs, but mainstream adoption would require coherence across memory subsystems, compiler support, and software ecosystems.

Advantages include larger natural data sizes, potential performance benefits for vectorizable workloads, and a longer horizon for addressing beyond 64-bit limits. Challenges include increased silicon area and power, compatibility issues with existing software, and diminishing returns for workloads not benefiting from wide widths. In cryptography, 128-bit security is standard for symmetric keys, and 128-bit blocks are used by algorithms like AES, but these aspects are distinct from the hardware data path and do not alone define 128bitisation.

Future prospects are uncertain and likely domain-specific, with gradual adoption in high-performance and specialized processors, while general-purpose consumer CPUs are more likely to support 128-bit vectors and types before full 128-bit general registers become common.