Intelligent Factories: MCUs with NPUs Enable AI-Driven Factory Automation

Advanced high-precision robot arm in an electronics factory: AI technologies are essential to factory automation. Image credit: Shutterstock

AI in Factory Automation: Key Challenges and Opportunities

One of the main challenges in factories is the increasing complexity of manufacturing processes, requiring sophisticated monitoring, control, and optimization systems. As product life cycles shorten and demand for customization grows, manufacturers are adapting to flexible and agile production methods that necessitate real-time data collection, analysis, and decision-making capabilities for efficient operations and quick responses to changing market needs.

Another challenge is maintaining product quality and consistency while minimizing defects and waste. Standard quality control methods rely on manual inspection and sampling, a time-consuming, error-prone, and ineffective process in detecting anomalies. Moreover, the sheer volume of data generated by equipment and sensors can overwhelm data processing and analytics systems, making it difficult to extract meaningful insights. 

There is also a need for integration between AI systems and existing automation infrastructure. Many factories utilize legacy equipment and proprietary protocols that were not designed with AI in mind. Retrofitting these systems can be a complex and costly endeavor requiring investments in hardware, software, and training. Standardization and interoperability issues between vendors and platforms also make it challenging to scale AI solutions across facilities or supply chains.

Despite these challenges, the potential benefits of AI in factory automation are too compelling to ignore. For example, AI-powered computer vision can automate product quality inspection tasks to detect defects with accuracy and speed. This capability not only improves quality and consistency, but also minimizes waste and re-work. 

With ML algorithms and models, AI also facilitates real-time monitoring and control of operations, predictive maintenance of equipment, and optimization of scheduling and resource allocation. For instance, algorithms can analyze sensor data to identify patterns and anomalies indicative of equipment failures. Maintenance teams can use this data to address problems before they escalate, reducing downtime and repair costs. 

AI can also facilitate the optimization of supply chain operations — from demand forecasting and inventory management to logistics. By analyzing historical data, market trends, and customer behavior, algorithms can predict future demand, enabling factories to adjust production volumes, optimize inventory, and reduce lead times. In warehouses, AI-powered robotics and autonomous guided vehicles (AGVs) can streamline picking, packing, and shipping processes, reducing labor costs while improving throughput. 

Perhaps most importantly, AI can enhance human-machine interactions in factory environments. With natural language processing and computer vision, AI can enable more intuitive and efficient communication between workers and machines, reducing training time and improving safety. 

The Role of Dedicated AI Hardware in Factory Automation

Dedicated AI hardware supports factory automation. Image credit: Shutterstock

To fully harness the capabilities of AI/ML in factory automation, dedicated AI hardware, such as MCUs are essential. However, while many solutions claim to be “AI-enabled” or “AI-supported,” they usually lack the necessary processing power and efficiency to handle complex industrial AI tasks — this is where a dedicated Neural Processing Unit (NPU) becomes indispensable. An NPU is a specialized processor designed for AI workloads. Unlike general-purpose CPUs, NPUs have architectures optimized for the unique computational requirements of AI/ML algorithms — they can perform massive parallel processing, handle large amounts of data, and execute complex neural network models with high efficiency. 

Running AI workloads on non-specialized hardware results in high power consumption and heat generation, which are problems in industries with strict energy and cooling requirements. Many factory automation tasks require real-time processing and decision-making, a challenging achievement without the fast and efficient processing capabilities of an NPU. Moreover, advanced AI algorithms and models often require specific hardware optimizations that are only available in dedicated NPUs, restricting the range of AI techniques that can be implemented. 

On the factory floor, an NPU obtains data from sensors, cameras, or other sources, and performs preprocessing steps, including normalization, feature extraction, and data augmentation. The pre-trained AI model is loaded onto the NPU, which executes it with its specialized architecture, performing matrix multiplications, convolutions, and other computationally-intensive operations. The NPU then generates predictions or decisions based on the input data and trained model, such as identifying defects in product images, predicting an equipment’s failure, or optimizing a production schedule. As new data emerges, the NPU fine-tunes the model using techniques like transfer learning or incremental learning, allowing it to adapt to changing conditions and to improve its performance over time. 

Alif’s MCUs Power Transformation in AI-Driven Factories

A Alif Balletto MCU. Image credit: Alif Semiconductor

From predictive maintenance and quality control to supply chain optimization and autonomous material handling, each application requires a combination of performance, power efficiency, wireless connectivity and flexibility. Harnessing the power of Arm® Ethos™ NPUs, Alif Semiconductor has developed the Ensemble® and Balletto™ families of AI/ML enabled MCUs which can meet the unique challenges of AI in factory automation. 

Alif Semiconductor’s MCUs combine high-performance processing and ultra-low power consumption, advanced connectivity, and comprehensive security features designed around a scalable multi-core architecture that incorporates Arm CPUs, NPUs, as well as Alif’s proprietary power management and sensor processing. This heterogeneous architecture enables Alif’s Ensemble and Balletto MCUs to handle diverse workloads of industrial processes, ranging from real-time sensor data processing and protocol conversion to complex AI inference and decision-making. And the integrated BLE and 802.15.4 functionality in the Balletto family also adds wireless communication capabilities.

Empowering Edge AI

Alif’s MCUs include Arm Ethos-U55 Neural Processing Units (NPU), which can perform up to 250 GOPS and is backed by a large tightly coupled memory (TCM). Neural network processing performance is up to two orders of magnitude faster than what a Cortex®-M4 processor can achieve for similar workloads, allowing significant improvements in efficiency and safety. Alif’s Ensemble and Balletto MCUs can intelligently run AI models that can be used to automate manufacturing operations such as assembly, packaging, and material handling, without compromising on power efficiency. 

Building on the success of the Ensemble and Balletto families, Alif Semiconductor has now shared that they are an early adopter of Arm’s just-released Ethos-U85 NPU. A crucial aspect of AI-enabled factory automation is the ability to work with a wide range of data types and models. The Ethos-U85 supports int8 weights and int8 or int16 activations, providing the flexibility to deploy and optimize various AI models, from compact, lightweight networks for Edge devices to large, complex models for centralized processing. 

In addition to performance and flexibility improvements, the Ethos-U85 NPU introduces features that directly address the complex requirements of AI in industrial settings. One such feature is its native support for transformer architecture networks, which have proven highly effective in tasks involving sequential data and long-range dependencies. Transformer networks are well-suited to automation applications such as predictive maintenance, where the ability to analyze time-series data and capture complex patterns is critical. 

Processing Enhancements

Alif’s Ensemble and Balletto MCUs can deliver the performance required for even the most demanding AI applications, such as real-time machine vision for quality control, predictive maintenance based on sensor fusion, and autonomous material handling using deep reinforcement learning. By bringing this level of processing power to the edge, Alif Semiconductor is enabling manufacturers to realize the full potential of AI, with smarter, more adaptive, and more efficient production systems.

Connectivity

A key advantage of Balletto is comprehensive wireless connectivity, which includes support for Bluetooth Low Energy, Matter, and Auracast. This enables integration with a wide range of sensors, actuators, and other endpoint devices in the factory environment, allowing for real-time data collection, analysis, and control. By leveraging the latest wireless standards and protocols, the Balletto family helps manufacturers create connected, responsive, and flexible production systems that can adapt to changing demands and optimize performance in real-time. 

Energy Efficiency

To address the issue of energy efficiency, Alif has equipped its MCUs with its proprietary aiPM™ technology. This autonomous power management system dynamically optimizes power consumption based on workload needs, ensuring that only the necessary components are active at any given time. With aiPM, Alif allows manufacturers to deploy AI-powered sensors, robots, and other devices that can run for longer periods on a single charge, minimizing maintenance costs and increasing operational efficiency.

Sensor Fusion Capabilities

Alif’s Ensemble and Balletto MCUs also provide advanced audio signal processing and hardware-accelerated sensor fusion capabilities. These features are critical for intelligent monitoring, control, and optimization in factory automation. For example, by analyzing machine sounds and vibrations, manufacturers can detect signs of equipment failure and schedule predictive maintenance. Similarly, by fusing data from sensors, such as temperature, humidity, and pressure, operators can understand their processes better, enabling them to improve product quality and reduce waste. 

Security

Security is another critical aspect of AI in factory automation, which Alif addresses with a holistic security architecture that includes an advanced secure enclave. This ensures the protection of sensitive data, intellectual property, and the integrity of AI models and algorithms deployed in factories. Manufacturers can deploy AI-powered solutions with confidence, knowing that their systems are protected against unauthorized access, tampering, and cyberattacks. 

Automated automobile factory. Image credit: Shutterstock

Conclusion

To realize the full potential of AI in factory automation, manufacturers require high-performance, energy-efficient hardware that can process vast volumes of real-time data under the challenging conditions of factory floors. Unlike general-purpose CPUs, NPUs offer the computational power, flexibility, and efficiency needed to drive AI in factories. With their scalable architectures, support for diverse data types and AI models, and hardware-level optimizations for key workloads, NPUs enable manufacturers to deploy AI solutions that can transform every aspect of their operations, from predictive maintenance and quality control to supply chain optimization and autonomous material handling. 

With high-performance processing, low power consumption, advanced connectivity, and security mechanisms, Alif Semiconductor leverages dedicated NPUs to create intelligent, connected, and secure AI solutions. 

For more information on Alif’s Ensemble and Balletto MCUs or product-specific inquiries, please visit the website.