Today, we’re welcoming back Scott from Ambiq, a familiar face around here and a big name in low-power tech. If you’re curious about where wearables and edge devices are headed, stick around. Scott’s here to walk us through Ambiq’s recent innovations and break down how AI’s emerging role will impact everything from design to data efficiency.
Ambiq’s Advances in Edge AI with Apollo 5
Scott provides an update on Ambiq’s progress, particularly focusing on the Apollo 5 processor, which targets Edge AI with significant efficiency improvements. Key features include the ultra-low power ARM Cortex-M55, designed to handle demanding neural network tasks with up to 5x faster processing than earlier models. Ambiq’s Apollo 5 family is now seeing adoption, with numerous customers beginning development, highlighting the surging demand for Edge AI capabilities in wearable tech and beyond.
The Complexities of Software Development in AI Hardware
Scott emphasizes that while the Apollo 5 is hardware-optimized for low-power AI, most development time is spent on software, particularly on adapting cloud-trained neural networks to function efficiently at the edge. Software optimization is essential, as translating complex AI models to compact, power-efficient processors presents unique challenges. Tools like TensorFlow Lite Micro and custom code have been employed, and Ambiq frequently collaborates with clients to fine-tune AI models for edge deployment.
AI Development Environments and Tools
The conversation shifts to discuss the tools and environments that support AI development, including Arm’s CMSIS-NN libraries and Google’s TensorFlow Lite Micro framework. Edge Impulse is noted for its developer-friendly platform, featuring a model zoo and a large community that facilitates the rapid development and deployment of edge AI applications. Scott highlights the democratization of AI development, where even non-electronics companies are building sophisticated AI solutions, such as AI-driven health detection in litter boxes and hospital-grade cough detection systems.
Small AI vs. Big AI: The Current Landscape and Future Prospects
Scott differentiates between “small AI” and “big AI.” Currently, most clients are engaging with small AI, using neural networks to replace traditional algorithms, like noise reduction in hearing aids or heart health monitoring in wearables. This incremental shift allows for performance improvements while avoiding major disruptions in design. However, big AI is on the horizon—applications that require large neural networks and significantly more memory and computational power. The industry anticipates challenges related to power consumption, memory constraints, and efficient data handling in these larger, more complex applications.
Memory Constraints and the Path Toward Big AI
The limitations of current microcontrollers (MCUs) in handling large AI workloads are discussed. While Ambiq’s processors support off-chip memory, big AI will demand even more from memory architectures, potentially creating bottlenecks. Scott mentions that Ambiq is preparing for this future by enhancing memory interfaces and exploring next-gen hardware to support high-memory AI applications. This approach allows customers to use off-chip memory, such as NAND and NOR flash, for larger datasets and neural network models.
AI’s Impact on Design and Engineering
The conversation turns philosophical as the hosts discuss the “coming storm” of AI’s integration into various aspects of design. As AI applications require more power, data, and heat management, design engineers will need to re-evaluate every system component to accommodate these demands. This “storm” signifies an industry-wide shift where designers must think beyond current specifications, preparing for a future where large AI applications drive new technical requirements.
Practical Examples: AI in Healthcare and Emergency Services
Scott shares real-world applications like Biostrap’s health monitoring wearables for firefighters. This device leverages neural networks to monitor critical health indicators, such as sleep quality, activity levels, and even stress, in high-risk jobs. The goal is not to detect minor, day-to-day stress variations but rather to identify long-term trends that could indicate serious health issues. These insights are especially valuable in emergency services, where both mental and physical health are crucial to effective performance.
Anticipating the Next Phase: Planning for AI’s Long-Term Integration
Scott underscores that AI integration requires careful planning, as big AI will introduce new technical challenges. Design engineers will need strategies to accommodate diverse memory demands and data requirements. He suggests that developers think of off-chip memory as an “off-planet” resource, underlining the importance of rethinking data storage and retrieval in AI applications.
The Boundless Future of AI
In closing, Scott emphasizes the exciting opportunities that AI brings, particularly in the medical and healthcare fields, where it could extend human lifespans by decades. He believes that if society can adapt to this technology thoughtfully, AI will enable groundbreaking advancements across industries. The conversation leaves listeners with the concept of “off-planet” thinking and the limitless potential of AI to transform design and engineering.
Check out all of ipXchange’s interviews with Ambiq’s partners at Embedded World 2024:
Apollo4 vs. Apollo5: Ambiq’s MCUs in (AI) action!
How Ambiq enabled the RAK11720’s added features
Creating stunning MCU-driven GUIs with Qt Group
When Ambiq meets energy harvesting – Live demo!