Technical Deep Dive
The Thus A1 chip represents a fundamental architectural departure from conventional digital signal processors (DSPs) and even standard AI accelerators. At its core is in-memory computing (IMC), a technique that performs analog computation directly within the memory array, eliminating the energy and time penalty of shuttling data between separate memory and processing units. This is critical for audio AI workloads: a typical neural network for speech enhancement might require 10-50 million multiply-accumulate (MAC) operations per second, and with traditional von Neumann architectures, data movement can account for over 90% of total energy consumption.
The Thus A1 integrates a custom SRAM-based compute-in-memory macro array, optimized for the low-bit-width (INT4/INT8) quantized models used in real-time audio processing. The chip achieves a power envelope under 10mW during active inference—low enough to be powered by a standard 40mAh earbud battery without sacrificing battery life. Latency is kept under 10 milliseconds, critical for real-time two-way communication.
Anker and Zhixin Technology developed a proprietary neural network architecture called VoiceFocusNet, which is distilled from a larger teacher model (likely based on a transformer or Conv-TasNet variant) down to roughly 2 million parameters. This model runs entirely on the Thus A1, performing three key tasks simultaneously:
- Speech Extraction: Separating the user's voice from background noise using a time-domain masking approach.
- Acoustic Context Classification: Identifying the noise environment (subway, street, office) to adapt filtering parameters in real time.
- Voice Reconstruction: Generating a clean, natural-sounding speech signal, avoiding the "tinny" artifacts common in traditional noise gates.
| Benchmark | Traditional DSP | Thus A1 (IMC) | Improvement |
|---|---|---|---|
| Power draw (active inference) | 25-40 mW | 6-9 mW | 4-6x lower |
| End-to-end latency | 15-25 ms | 6-9 ms | 2-3x faster |
| Model capacity (parameters) | <100K (rule-based) | 2M (neural) | 20x more |
| Noise reduction (SIIB, 0-1) | 0.65 | 0.89 | +37% |
Data Takeaway: The Thus A1 achieves a 4-6x power reduction and 2-3x latency improvement over traditional DSPs while running a neural network 20x larger. This is the core enabler of the Guinness record—not just better filtering, but fundamentally different AI-driven speech understanding.
A relevant open-source reference point is the speechbrain repository (GitHub, 8k+ stars), which provides a toolkit for building speech processing models. However, no open-source project currently matches the Thus A1's power efficiency, as IMC requires custom silicon. The TinyML ecosystem (TensorFlow Lite Micro, 20k+ stars) is the closest software analog, but it still relies on conventional MCU architectures.
Key Players & Case Studies
Anker Innovations is the prime mover here. Known primarily for charging accessories and power banks, Anker has been quietly building a formidable audio division under its Soundcore brand. The Liberty 5 Pro series is a direct assault on the premium earbud market dominated by Sony (WF-1000XM5), Apple (AirPods Pro 2), and Bose (QuietComfort Ultra Earbuds). Anker's strategy is differentiation through edge AI, not just hardware specs.
Zhixin Technology (知存科技) is the silicon partner. Founded in 2017, Zhixin is one of the few companies globally with a commercial in-memory computing chip. Their WTM-8 series, used in the Thus A1, is fabricated on a 28nm process—mature enough for cost-effective mass production. Zhixin has raised over $100M in funding from investors including Xiaomi and Sequoia Capital China, and their roadmap includes 12nm and 7nm nodes for higher-performance applications.
| Competitor | Chip | Architecture | Power (active) | Application |
|---|---|---|---|---|
| Anker/Zhixin | Thus A1 | SRAM IMC | 6-9 mW | Audio AI |
| Apple | H2 | DSP + NPU | ~15 mW | Audio + ANC |
| Qualcomm | S5 Gen 2 | DSP + Hexagon NPU | ~20 mW | Audio + voice |
| Sony | Integrated | Custom DSP | ~18 mW | Audio + ANC |
Data Takeaway: The Thus A1's power advantage is not marginal—it is 2-3x better than the closest competitor (Apple H2) for the specific task of neural audio processing. This allows Anker to run larger models without battery life penalties.
Anker CEO Yang Meng has publicly stated that the Thus A1's capabilities will extend beyond call noise reduction to active noise cancellation (ANC) and audio quality enhancement. This is a strategic pivot: instead of using off-the-shelf ANC chips from Qualcomm or Analog Devices, Anker can now differentiate through AI-driven ANC that adapts to the user's ear shape and environment in real time.
Industry Impact & Market Dynamics
The Thus A1's arrival signals a broader shift: edge AI is finally becoming practical for power-constrained wearables. The global TWS (True Wireless Stereo) earbud market is projected to reach $150 billion by 2027, with AI features being the primary differentiator. Currently, most "AI" features in earbuds are cloud-dependent (e.g., real-time translation) or rely on simple rule-based algorithms. The Thus A1 brings large-model inference entirely on-device, enabling privacy-preserving, low-latency features.
This has implications beyond audio. In-memory computing is being explored for vision (smart glasses), health monitoring (wearable ECG), and always-on voice assistants. The Thus A1 is a proof point that IMC can deliver commercial-grade performance at consumer price points.
| Market Segment | 2024 Value | 2027 Projected | CAGR | AI Chip Penetration (2027) |
|---|---|---|---|---|
| TWS Earbuds | $85B | $150B | 15% | 40% |
| Wearable AI Chips | $2.5B | $12B | 48% | — |
| In-Memory Computing | $0.8B | $8B | 75% | — |
Data Takeaway: The IMC chip market is expected to grow 10x in three years, driven by edge AI wearables. Anker's Thus A1 is an early mover in this high-growth segment.
However, the competitive response will be swift. Apple is reportedly developing its own IMC chip for the next-generation AirPods, and Qualcomm is investing in analog compute-in-memory for its Snapdragon X series. Anker's first-mover advantage may be limited to 12-18 months.
Risks, Limitations & Open Questions
1. Generalization: The VoiceFocusNet model is trained on a specific set of noise profiles. Edge cases—like a crying baby or a jackhammer—may still degrade performance. Anker needs to demonstrate robustness across a wider range of real-world scenarios.
2. Scalability: The Thus A1 is a dedicated audio chip. Can Anker extend the same architecture to other product categories (e.g., smart glasses, speakers) without a complete redesign? The chip's SRAM macro array is optimized for audio model sizes; vision models require 10-100x more memory.
3. Supply Chain: Zhixin Technology's 28nm process is mature, but moving to 12nm or 7nm for future iterations will require significant capital expenditure and yield management. Any delays could cede advantage to Apple or Qualcomm.
4. User Perception: The Guinness World Record is a powerful marketing tool, but consumers may not perceive a dramatic difference in everyday use. Anker must ensure the AI features are demonstrably better, not just benchmark-beating.
5. Battery Life Trade-off: While the Thus A1 is efficient, running continuous AI inference still consumes power. Anker claims 10 hours of talk time—competitive, but not class-leading. The next generation must improve this.
AINews Verdict & Predictions
Prediction 1: The Thus A1 will become a platform, not a one-off chip. Anker will release a developer SDK for third-party audio brands to integrate the chip, creating a licensing revenue stream. This mirrors Qualcomm's Snapdragon Sound strategy but with a superior AI core.
Prediction 2: In-memory computing will become the standard for wearable AI within 3 years. The Thus A1 proves that IMC can deliver the power/latency/performance trifecta. Expect Apple, Samsung, and Google to announce their own IMC chips for wearables by 2026.
Prediction 3: The Guinness record will trigger a wave of similar certifications. Sony and Bose will likely submit their next-gen earbuds for comparable tests, leading to a "benchmark war" in audio AI. This is good for consumers but may lead to over-optimization for test conditions.
Prediction 4: Anker's stock and brand valuation will see a significant boost. The company has successfully pivoted from "accessory maker" to "AI hardware innovator." This will attract new investors and partnerships, particularly in the automotive and smart home sectors where Anker already has a presence.
What to watch next: The Liberty 5 Pro Max's teardown and independent benchmarks. If third-party tests confirm the Guinness results, Anker will have a genuine breakthrough. Also, watch for Zhixin Technology's next chip—if it targets vision or health sensing, the IMC revolution will expand beyond audio.