Portable CBRN Control System

Artificial Inteligence (AI) and Machine Learning (ML)

We have working prototypes built on AI technologies that we are excited to share with customers who are interested in object recognition, object segmentation, neural networks, and phrase detection. DE is focused on Artificial Intelligence (AI) and Machine Learning (ML) at the edge or leaf area of Internet of Things (IoT); that is, the sensor and control node embedded hardware and firmware level. One tool we use is tensor flow-based learning to create custom tailored AI solutions. Our solutions can have the ability to still work offline or off-grid, where internet may not be available or may be intermittent. Making our solutions well suited for portable medical device applications. Our background in medical device development, mobile devices, power supply design, and highly accurate sensors and analog design positions us well to be an excellent partner for development of AI/ML based low level portable hardware devices.

Let us know if you have a need for a mobile AI product in your industry.

Our Process

Let’s Talk About Your Project

Fill out our form, and our New Project Team would be happy to meet online for 30 minutes to discuss your project and share our capabilities.

We use your information to learn more about your project and to send occasional emails with an easy opt-out (we're not fans of inbox clutter either). And your data stays private and unsold, as it should.


Frequently Asked Questions

What does AI and machine learning integration mean in the context of embedded electronics?

AI and ML integration in embedded electronics means deploying trained inference models directly on edge hardware — microcontrollers, DSPs, FPGAs, or application processors — rather than in the cloud. The hardware must be selected to support the compute requirements of the model (matrix multiplications, convolutions), and the firmware must manage memory, power, and real-time scheduling alongside the inference engine. DE Design Works designs the hardware-software co-design that makes on-device AI practical for industrial and medical products.

What hardware platforms are suitable for running AI and ML inference on embedded devices?

Hardware selection depends on model complexity and power budget. Options range from Cortex-M microcontrollers with TensorFlow Lite Micro for keyword detection or simple classifiers, to dedicated NPUs (Neural Processing Units) like the NXP eIQ platform or Syntiant NDP processors for audio/vision inference, to embedded GPUs and FPGAs for more demanding models. DE Design Works evaluates model requirements (FLOPS, memory footprint, latency target) against candidate hardware platforms and selects the lowest-cost option that meets the specification.

What are the biggest challenges in deploying machine learning models on embedded hardware?

The three dominant challenges are: (1) Memory constraints — ML models can be megabytes in size, but many MCUs have only kilobytes of SRAM, requiring quantization, pruning, and architecture optimization; (2) Power budget — inference on a neural network consumes significant energy, requiring careful duty-cycling to meet battery life targets; and (3) Data pipeline design — collecting, preprocessing, and feeding sensor data into the model at the right rate while the rest of the firmware runs concurrently. DE Design Works addresses all three through co-design of the model architecture and hardware platform.

Can ML models trained in the cloud be deployed on resource-constrained embedded devices?

Yes, but they must be optimized for embedded deployment. Techniques include post-training quantization (converting float32 to int8, reducing model size by 4×), pruning (removing low-weight connections), and knowledge distillation (training a smaller student model from a larger teacher). Frameworks like TensorFlow Lite, Edge Impulse, and NXP eIQ automate much of this optimization pipeline. DE Design Works works with data science teams to ensure models are designed for deployability from the start — not as an afterthought.

How does DE Design Works approach AI/ML integration for industrial electronics products?

DE Design Works follows a hardware-software co-design approach: first define the inference task and latency/accuracy requirements, then select the processor architecture that meets those requirements at the lowest power and cost, then develop the firmware inference pipeline alongside the application software. For industrial products, DE Design Works also validates model accuracy under real-world sensor noise and environmental variation — lab accuracy rarely matches field accuracy without this step.