Neural Engine patent trend
- PatSnap Eureka published a patent analysis tracing Apple's Neural Engine evolution across A‑series and M‑series chips. - The analysis maps Neural Engine capability growth from about 0.6 TOPS in early designs to roughly 38 TOPS in newer architectures. - The patent trajectory indicates stronger on‑device AI acceleration and growing integration of neural compute into SoC roadmaps (x.com).
A Neural Engine is the part of a chip built to run artificial intelligence math, the way a graphics processor is built to draw images. Patent analysis published by PatSnap says Apple’s designs moved from about 0.6 trillion operations per second in the A11 Bionic in 2017 to 38 trillion operations per second in the M4 generation. (patsnap.com) Apple introduced the first Neural Engine with the A11 Bionic in the iPhone X on September 12, 2017, calling it a dual-core design that could handle 600 billion operations per second for features including Face ID and Animoji. By May 7, 2024, Apple said the M4 chip’s Neural Engine could reach 38 trillion operations per second, or 60 times the speed of that first version. (apple.com 1) (apple.com 2) The steps in between were large and uneven. Apple said the A14 Bionic in 2020 reached 11 trillion operations per second, the A15 in 2022 reached 15.8 trillion, and the M1 brought an 11-trillion-operations-per-second Neural Engine to the Mac in November 2020. (apple.com 1) (apple.com 2) (apple.com 3) PatSnap’s review tracks patents across both Apple’s A-series iPhone chips and M-series Mac chips, which is one way to watch how a company’s chip priorities shift before products arrive. Its write-up says Apple filed more than 29 Neural Engine-related patents from 2018 through 2025. (patsnap.com) Apple’s own software stack has been built around that hardware. The company’s Core ML tools say apps can use the central processor, graphics processor, and Neural Engine together to run machine learning while reducing memory use and power draw. (developer.apple.com) That hardware push now sits under Apple’s broader artificial intelligence strategy. When Apple introduced Apple Intelligence on June 10, 2024, it said many requests would run on device first, with larger jobs shifting to Private Cloud Compute on Apple silicon servers only when needed. (apple.com) Apple’s researchers have also described the Neural Engine as a target for newer model types, including Transformers, the model family behind many modern chatbots and text generators. In a 2021 research post, Apple said the first-generation Neural Engine shipped in the A11 chip and outlined methods for running Transformer models efficiently on that hardware. (apple.com) The patent trail does not prove what Apple will ship next, but it does line up with the products already on the market: more of the company’s AI work is moving onto the chip itself, and the Neural Engine is taking a larger share of that load. (patsnap.com) (apple.com)
