Self-developed SNN Neuromorphic Chip, Acting as the 'Upstream Brain' of Medical Devices, 'Mien Technology' Raises Tens of Millions in Funding | 36Kr Exclusive 自研SNN类脑芯片、做医疗设备的“上游大脑”，「米能科技」获数千万元融资｜36氪首发

TL;DR

• Minneng Technology has developed a proprietary Spiking Neural Network (SNN) neuromorphic chip designed specifically for medical devices, addressing the limitations of traditional MCU+ANN architectures in processing continuous physiological signals.
• The core innovation is an event-driven architecture that activates computation only when specific physiological events (e.g., ECG anomalies) occur, enabling ultra-low power consumption and 7x24-hour monitoring without overheating or battery drain.
• The company employs a "dual-track" commercialization strategy: rapid deployment of standardized modules for consumer health products to generate cash flow, while simultaneously building high-barrier clinical validation systems for serious neuro-medical applications like chronic pain management and sleep regulation.
• They offer a five-tier delivery model ranging from pure hardware supply (L1) to full regulatory support and small-batch manufacturing assistance (L5), catering to both established medical device giants and clinician-founded startups lacking hardware expertise.

Why It Matters

This article highlights a critical shift in AI hardware design: moving away from general-purpose deep learning models toward specialized, bio-inspired architectures optimized for real-time, low-power signal processing in safety-critical environments. For AI practitioners, it demonstrates the value of co-designing algorithms with hardware constraints—specifically how event-based computing can solve the latency and energy bottlenecks plaguing wearable health tech. For the industry, it underscores the growing importance of "upstream" foundational platforms rather than just end-user applications, suggesting future winners may be those who control the sensor-to-signal pipeline itself.

Technical Details

• Neuromorphic Core: Uses analog-digital hybrid SNNs mimicking biological neuron sparsity and asynchronous firing patterns, contrasting with conventional artificial neural networks that process data continuously regardless of relevance.
• Physiological Pulse Encoding Mechanism: Implements custom encoding schemes tailored to EEG, ECG, EMG waveforms; triggers computational wakefulness exclusively upon detecting feature mutations in biosignals rather than fixed sampling intervals.
• Three-Layer Hardware Stack:
  1. Multi-modal sensory array supporting dry/half-dry electrodes for capturing weak bioelectric/optical signals;
  2. Event-driven neuromorphic processor performing sparse filtering and state assessment;
  3. Programmable control unit enforcing multi-layered hardware-level safety rules before any intervention output.
• End-to-End Integration: Proprietary toolchains, standardized test datasets, and compliance frameworks eliminate third-party component dependencies, reducing integration friction for OEM partners.

Industry Insight

The rise of personalized medicine and aging populations will increasingly demand long-term, unobtrusive home-based monitoring solutions—making energy-efficient edge AI not optional but essential. Companies investing now in domain-specific neuromorphic infrastructure (like Minneng’s closed-loop sensing-computation-control stack) are positioning themselves as indispensable suppliers to legacy medtech firms struggling to modernize their product lines. Meanwhile, clinicians-turned-founders should prioritize partnerships over vertical integration if they lack semiconductor R&D capacity; tiered service models like L3–L5 lower entry barriers significantly. Finally, regulatory approval timelines remain a major hurdle—early collaboration with hospitals on clinical trials (as noted in their “serious medicine” track) could accelerate market access by validating both efficacy and safety concurrently with engineering development.

TL;DR

• 米能科技完成数千万元融资，专注于自研SNN类脑芯片作为医疗设备的底层算力平台。
• 其核心技术为脉冲神经网络（SNN），通过事件驱动机制实现超低功耗和实时生理信号处理。
• 公司构建了“感知—计算—调控”三层硬件闭环体系，提供从芯片到注册支撑的五级阶梯交付方案。
• 商业化采取双轨策略：消费医疗快速落地产生现金流，严肃医疗构筑高壁垒长期增长。
• 已覆盖睡眠调节、运动康复等场景，并与三甲医院共建临床验证体系。

为什么值得看

本文揭示了AI在医疗硬件领域的关键范式转移——从通用ANN转向原生适配生理信号的SNN架构，解决了可穿戴设备续航与延迟的核心痛点。对于AI从业者而言，这展示了端侧异构计算在垂直行业的深度落地路径；对医疗设备厂商而言，提供了绕过底层研发瓶颈、加速产品合规上市的标准化解决方案参考。

技术解析

1. SNN类脑内核设计：自研数模混合脉冲神经网络，复刻人脑异步工作逻辑，仅在EEG/ECG/EMG等波形出现特征突变时唤醒算力单元，无异常时进入深度休眠，从根本上解决连续监测的功耗问题。
2. 三层硬件闭环架构：包含多模态生理感知阵列（兼容干电极微弱信号采集）、事件驱动类脑内核（低功耗筛选与状态评估）、医疗级可编程调控单元（内置硬件安全防火墙规避算法黑盒风险）。
3. 专属编码与算法库：内置针对生理波形的脉冲编码机制及非开源的专属算法库，配合硬件安全逻辑形成完整知识产权壁垒，区别于通用ANN二次改造方案。
4. 开发工具链支持：统一开发工具链、标准化测试数据集和安全管控规则，使厂商无需多厂商联调即可直接嵌入终端，大幅缩短研发与医疗器械注册周期。
5. 分级交付体系：L1-L5五级方案覆盖从纯芯片供应到全套注册支撑的全类型客户，尤其赋能缺乏硬件能力的临床初创团队。

行业启示

1. 医疗算力底层重构趋势：传统MCU+ANN架构难以适配人体连续弱信号，原生事件驱动的类脑计算将成为下一代可穿戴及植入式设备的标准底座，推动硬件产业从“通用算力”向“生理原生算力”范式迁移。
2. 软硬分离生态加速成熟：上游芯片企业通过标准化模组和全链路服务（含注册支撑）降低下游整机厂门槛，形成“上游底层平台+下游终端应用”的分工模式，利好专注临床需求的轻资产创业团队。
3. 消费与严肃医疗双轮驱动：消费级场景快速验证平台通用性并产生现金流，反哺严肃医疗的高壁垒临床验证；未来随着无创调控技术管线拓展，慢病长效干预市场将释放巨大增量空间。