On Edge: Why the IoT Marketplace Needs a New Class of Silicon

Legacy silicon is struggling to keep pace with the Internet of Things (IoT). As AI steadily moves from the cloud to the Edge, conventional silicon architectures are proving to be inadequate for the demands of the next generation of AI-enabled IoT. Legacy MPUs, after all, lack the capabilities needed for efficient Edge processing. They lack the AI acceleration, real-time processing, and power efficiency needed to handle complex, data-intensive workloads outside the cloud. A new class of AI-native chips is needed.

Key Drivers Behind this Demand

Several powerful market forces are accelerating this shift in demand for a new class of AI-native silicon at the Edge. The limitations of cloud-based AI are becoming clearer, and both consumers and businesses are demanding solutions that are more private, cost-effective, responsive and energy-efficient. 

1. The Rise of AI at the Edge: Large Language Models (LLMs) are rapidly becoming smaller and more efficient, making it practical to run them directly on local devices instead of relying on the cloud.
2. Privacy: Customers are increasingly concerned about their personal data being sent to the cloud for processing. Keeping data on the device is a powerful way to address these privacy fears.
3. Cost: There is a growing awareness of both the direct financial costs of using cloud AI services and the indirect environmental costs.
4. Latency: For applications that require real-time responses, the delays inherent in sending data to the cloud and back are unacceptable.
5. Power Consumption: Energy efficiency is becoming an important requirement. The upcoming EU Ecodesign Regulations (2027) will impose strict maximum power consumption limits on household appliances and smart home devices. This reflects a global shift, with North America and the Asia-Pacific region expected to follow with similar restrictions, driving a worldwide trend in energy efficiency. 

The Shortfalls of Conventional Silicon

The silicon currently used in many IoT devices wasn’t designed for this new, AI-driven reality. It suffers from significant technical and business challenges that make it a poor fit for the future.

Technical Inadequacies

The existing silicon architectures are falling short in five critical areas: 

1. Lack of AI-Native Features: Legacy silicon often lacks support for crucial AI operations, such as transformers, which are a fundamental building block for many modern mathematical operations at the Edge. While transformer support has existed in high-end, expensive devices, bringing this capability to cost-effective, single-digit-dollar silicon marks a paradigm shift.
2. Limited Integration: Current chips lack the deep integration of specialized accelerators, which limits their ability to handle future AI workloads.  
3. Cost-Prohibitive Design: Much of the existing silicon is repurposed from other markets, like automotive. These chips come with expensive, unnecessary features for IoT applications, such as automotive-grade IP. Some vendors even use a "binning" strategy, where chips that don’t qualify for automotive standards are pushed toward the consumer market, rather than designing silicon specifically for IoT needs.
4. Sub-Optimal Performance and Power: Existing solutions often feature inadequate MCU cores and consume more power for IoT use cases. The metric for success is evolving to performance per watt, per dollar, and AI capability is now an inherent part of that performance calculation.  
5. Constrained Memory Options: Alternative suppliers often limit designers to LPDDR4 memory, whereas IoT AI processors demand flexibility with support for DDR3, DDR4, and LPDDR4, giving customers options to navigate supply chain availability and cost. 

Business and Ecosystem Challenges

Beyond the technical limitations, the business practices of large semiconductor vendors create additional risks for IoT device makers:

1. Vendor Lock-In: Large silicon vendors are acquiring Edge AI software providers. These vendors then bundle the proprietary software with their hardware, limiting customers into closed ecosystems that stifle innovation and portability.  
2. Supply Chain Risk: The COVID-19 pandemic highlighted weaknesses in the supply chain. Large semiconductor vendors often prioritize their most valuable customers, leaving other OEMs without adequate support. This experience has motivated a widespread need for a reliable second-source strategy. 

Defining the New Class of Silicon

To overcome these challenges, the market needs a new class of silicon defined by a fundamentally different approach.

Core Characteristics

This new silicon is:

• Purpose-Built: It is AI-native, multimodal silicon designed specifically for both consumer and industrial IoT applications, without the baggage of non-IoT IP.
• Optimized for a Better Metric: It delivers the best performance per watt, per dollar, a metric that inherently includes AI capabilities.
• Future-Ready: It is designed to be flexible and extensible, ready to support future AI developments, and provide customers with an insurance policy for what’s next in AI. For instance, the SL2610 product line is designed to meet and exceed the 2027 EU Ecodesign Regulations. 

Design and Platform Advantages

• Pin-to-Pin Compatibility: An entire family of products is pin-to-pin compatible, which allows customers to easily scale features up or down across multiple products without redesigning their hardware. This dramatically improves their time-to-market.
• Open Hardware and Software: The new approach combines flexible hardware with a truly open-source software experience. This allows silicon innovation to keep pace with rapid software innovation. By using open standards, such as the MLIR-compliant IREE compiler and runtime, this approach democratizes AI development, moving away from proprietary vendor-specific tools.
• Strong Ecosystem Partnerships: Collaboration with industry leaders gives developers access to this open approach and helps build a large ecosystem built on best practices. 

Real-World Applications 

This new class of silicon unlocks tangible, real-world applications that were previously either too impractical or too costly to implement.

• In the Smart Home: Imagine having a natural conversation with your dishwasher. You ask, "Why aren’t some of my dishes getting washed?" and the appliance tells you the filter is blocked, because all the necessary information and processing power is right there on the device. Or picture a thermostat with a rich, animated 3D graphical interface that enhances the user experience, a feature not possible with older, 2D GPU-enabled silicon.
• In the Factory: In industrial settings, this silicon enables cost-effective, vision-based Edge AI systems that are deployed on assembly lines to automatically detect defects or contaminants, replacing slow and expensive manual inspection. It is also ideal for predictive maintenance, such as analyzing time-series data to monitor the air sealed into confectionery packaging and predict failures before they disrupt production. 

Building the Future of IoT, Together

The convergence of market demands for privacy, cost-efficiency, and real-time responsiveness, combined with the limitations of legacy hardware, makes one thing clear: a new class of AI-native, cost-effective, and open silicon is essential for the next wave of IoT innovation. We invite developers to participate in this new, open development model. Let’s build the future of Edge AI applications together.