Edge and Embedded AI Hardware: Jetson, NPUs and Mobile SoCs
For embedded, robotics and device engineers who must choose and deploy AI silicon under tight power, thermal, cost and latency budgets, from Jetson modules to laptop and phone NPUs and discrete edge accelerators.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
Edge AI is constrained by watts, heat, memory bandwidth and bill-of-materials cost, and the datasheet number that marketing leads with (TOPS) is rarely what limits a deployed model. This course teaches engineers to read edge AI silicon critically and to carry a model from a training framework onto real hardware. It starts with the architectures: NPUs as fixed-function or programmable matrix engines, integrated GPUs, DSPs and CPU vector units, and how memory bandwidth and on-chip SRAM bound performance for vision and language models.
It then surveys the hardware current in October 2026: NVIDIA Jetson Orin modules and the Blackwell-based Jetson Thor family; PC NPUs in Qualcomm Snapdragon X2, Intel Core Ultra and AMD Ryzen AI processors, and the Copilot+ PC NPU requirement; Apple M-series and A-series Neural Engines and GPU neural accelerators; mobile SoCs from Qualcomm and MediaTek; and discrete and microcontroller-class accelerators such as Hailo and Google's Edge TPU. The second half is practical: INT8 and INT4 quantization and its accuracy cost, toolchains (TensorRT, ONNX Runtime execution providers, Qualcomm AI Runtime, OpenVINO, Core ML, LiteRT), power modes and thermal design, carrier boards and camera and sensor I/O, fleet updates and security, and a worked selection for a robotics vision system and an on-device assistant. TOPS figures are vendor peak ratings and are labeled with precision and sparsity where the vendor states them. The course ends with an automated exam and a Wirewalk Institute certificate of completion and assessment; it is not a vendor certification.
What you will be able to do
Explain how NPUs, integrated GPUs, DSPs and CPU vector units differ in architecture, programmability and efficiency for edge AI
Interpret TOPS ratings by precision, sparsity and sustained power, and estimate real throughput from memory bandwidth and model size
Compare NVIDIA Jetson Orin and Thor modules, PC and mobile SoC NPUs, and discrete edge accelerators for a stated workload and power budget
Quantize a model to INT8 or lower, measure the accuracy cost, and choose the toolchain and runtime for the target silicon
Plan power modes, thermal design and enclosure constraints so a device sustains its required performance
Specify carrier board, camera, sensor and storage I/O for an embedded AI system
Produce a justified hardware selection and deployment plan for an edge AI product, including updates and security
Syllabus
6 modules · 20 lessons · about 12 hours
1
Module 1: Edge AI Architectures
The compute engines inside edge silicon and what actually bounds their performance.3 lessons
Why the Edge Is Different: Power, Thermal, Latency and Cost Budgets
NPUs, Integrated GPUs, DSPs and CPU Vector Units
Reading TOPS Honestly: Precision, Sparsity, Sustained Power and Memory Bandwidth
2
Module 2: NVIDIA Jetson
The Jetson Orin and Thor families, their power modes and software stack.3 lessons
Jetson Orin: Nano, NX and AGX Modules and Power Modes
Jetson Thor: Blackwell at the Edge for Robotics and Physical AI
JetPack, TensorRT and DeepStream on Jetson
3
Module 3: PC and Mobile SoC NPUs
NPUs in laptops and phones, and how software reaches them.4 lessons
PC NPUs: Snapdragon X2, Intel Core Ultra, AMD Ryzen AI and the Copilot+ PC Bar
Apple Silicon: Neural Engine, GPU Neural Accelerators and Unified Memory
Mobile SoCs: Snapdragon and Dimensity for On-Device Language and Vision Models
Reaching the NPU: ONNX Runtime, Qualcomm AI Runtime, OpenVINO, Core ML and LiteRT
4
Module 4: Discrete and Microcontroller-Class Accelerators
Add-in accelerators and tiny ML silicon for cameras, gateways and sensors.3 lessons
M.2 and PCIe Edge Accelerators: Hailo and Similar Devices
Edge TPU, Industrial SoCs and Microcontroller NPUs
Choosing Between an SoC NPU, a Jetson Module and an Add-In Accelerator
5
Module 5: Getting Models onto Devices
Quantization, compilation and measurement on the target hardware.3 lessons
Quantization for Edge: INT8, INT4 and Mixed Precision
Compilation, Operator Support and Fallbacks
Profiling on Device: Latency, Throughput and Energy per Inference
6
Module 6: Designing and Shipping Edge Systems
Thermal, I/O, fleet and security engineering, and a worked hardware selection.4 lessons
Thermal Design and Power Budgets: Sustained Versus Burst Performance
Carrier Boards, Camera and Sensor I/O and Storage
Fleet Updates, Secure Boot and Model Protection
Worked Selection: Robotics Vision System and On-Device Assistant
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by NVIDIA, Qualcomm, Apple, Intel, AMD, MediaTek, Arm, Hailo, Google, NXP, or Microsoft. Product and company names are trademarks of their respective owners.
A plain-English, one-hour walk through an AI data center: what a model asks of hardware, what is inside a GPU server, how GPUs talk over NVLink, InfiniBand and Ethernet, where the data lives, and why AI racks need so much power and cooling.
A plain-language tour of the hardware that runs AI: why AI is limited by compute, memory bandwidth and data movement; what GPUs, HBM and precision formats are; how servers, networks, storage, power and cooling fit together; and how to size a simple inference or training system and weigh cloud against buying.
An engineer's tour of the modern data center GPU: how SMs and CUs, warps and wavefronts, tensor and matrix cores and the on-chip memory hierarchy work; what FP64 through FP4 and sparsity really buy you; how HBM, PCIe, NVLink, Infinity Fabric, UALink and CXL move data; and how power, cooling and reliability features shape real deployments. Reviewed against vendor sources as of October 2026.