Wireless earbuds run on batteries smaller than a dime, which means every milliwatt counts. That constraint has kept advanced AI features like real-time noise suppression tethered to smartphone processors or cloud servers. Anker Innovations is betting it can change that with a custom chip called “Thus,” a compute-in-memory design that runs neural networks directly inside the earbud rather than offloading the work elsewhere.
The chip, first detailed by German technology outlet Heise and covered by related publications, ditches the conventional approach of pairing a processor with separate memory. Instead, Thus stores AI model weights inside NOR-flash memory cells and performs the math right there in the memory array. That eliminates the constant back-and-forth data shuttling between processor and memory, which is one of the biggest power drains in small devices. According to Heise’s reporting, the result is roughly 5 billion operations per second on just a few milliwatts of power, enough to handle models with several million parameters in real time.
Why compute-in-memory matters for earbuds
The core problem Thus targets has a name in chip design: the “memory wall.” A peer-reviewed paper published in Frontiers in Science explains how conventional architectures waste enormous energy fetching data from separate memory chips. The paper details how compute-in-memory (CIM) designs sidestep this by performing multiply-accumulate operations, the mathematical backbone of neural networks, directly where the weights are stored. The energy savings can be dramatic, which is exactly what you need when your entire power budget fits inside a coin-cell-sized battery.
Many earbud chipmakers already ship dedicated neural accelerators, but most rely on SRAM or DRAM to hold model weights. That requires extra memory chips, more board space, and more power. A NOR-flash CIM design like Thus could, in theory, merge storage and compute onto a single die, shrinking the hardware footprint and simplifying the internal layout of an earbud. If the implementation works as described, midrange earbuds could offer the kind of always-on AI noise suppression currently reserved for premium models, without halving battery life in the process.
The idea is not brand new, but the application is
Merging computation with memory has commercial precedent stretching back decades. In the 1990s, Sun Microsystems and Mitsubishi shipped a product called 3DRAM that combined DRAM with logic circuitry to accelerate graphics workloads. What sets Anker’s effort apart is the target: NOR-flash-based AI inference inside consumer audio hardware, a category where sealed enclosures, tiny batteries, and strict thermal limits make traditional processor-plus-memory designs impractical for running neural networks continuously.
The manufacturing infrastructure to support this kind of chip already exists at commercial scale. GlobalFoundries and Microchip Technology have announced that Microchip’s 28 nm SuperFlash embedded flash solution, branded GF 28SLPe, is in production. That platform targets low-power embedded devices requiring fast access times and reliable non-volatile storage, domains that overlap with what Anker describes for Thus. Anker has not publicly confirmed which foundry or flash process it uses, but the availability of production-ready 28 nm embedded flash platforms shows this class of chip is no longer confined to research labs.
What we still do not know
As of late April 2026, no official Anker press release, datasheet, or white paper on Thus has surfaced publicly. The technical claims, including the operations-per-second figure, parameter count, and power draw, come from Heise’s reporting rather than from documentation that outside engineers can independently audit. Without published benchmarks, it is difficult to compare Thus head-to-head with established earbud-class silicon from Qualcomm or MediaTek.
Several practical questions remain open. Anker has not named a foundry, disclosed partner brands, or announced specific device launch dates. Whether Thus will appear first in Anker’s own Soundcore-branded earbuds, in third-party products, or both is unknown. A broader roadmap hinting at smart home devices, wearables, or hearing aids remains speculative until backed by concrete product announcements.
Technical unknowns also linger. The Frontiers in Science paper validates the general energy-saving logic of CIM architectures but does not evaluate Anker’s specific NOR-flash implementation. Flash endurance is one concern: if noise-suppression models need periodic weight updates, repeated writes could stress the memory cells over time. Aggressive quantization, a common technique for shrinking models to fit tiny chips, could degrade audio quality if not carefully tuned. No teardown reports or independent lab measurements have appeared to confirm the chip’s internal architecture or real-world performance.
Software tooling is another blind spot. It is unclear whether Thus is a closed, pre-trained noise-suppression engine or whether Anker plans to release an SDK for third-party developers to deploy custom models. Without documentation on supported frameworks or update mechanisms, the chip’s flexibility relative to conventional embedded-AI accelerators is hard to gauge.
What this means for buyers and the industry
For consumers, the practical takeaway is straightforward: if Thus delivers on its reported specs, the next generation of wireless earbuds, potentially including Anker’s own Soundcore line, could run sophisticated AI-driven call clarity and noise cancellation without the battery penalty that currently limits those features. That would be a meaningful upgrade for anyone who relies on earbuds for work calls or noisy commutes.
For the broader chip industry, Anker’s move signals that consumer electronics brands are no longer content to buy off-the-shelf silicon. Designing custom compute-in-memory hardware is a significant investment, and the fact that a company best known for charging cables and Bluetooth speakers is making that bet suggests the competitive pressure around on-device AI in wearables is intensifying.
The evidence available right now sits in a credible but incomplete middle ground. Heise’s technical reporting is detailed and attributed, peer-reviewed research supports the underlying science, and the embedded flash supply chain is mature enough to manufacture this kind of chip at scale. What is missing is the primary proof: an Anker-published datasheet, third-party benchmarks, or an independent teardown. Until those arrive, Thus is best understood as a technically grounded and genuinely interesting development in low-power AI, one whose real-world impact will only become clear once commercial devices ship and reviewers can measure what the chip actually does inside a sealed earbud.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
