DBATU · Third Year B.Tech Artificial Intelligence & Allied · NEP 2020
Fundamental Edge AI
26AFAIPC603 · Semester VI · 3 credits · Continuous assessment 20 + mid-semester 20 + end-semester 60
Unit 1: Heterogeneous Systems
5 hoursSystem-on-chip architectures that combine application processors and microcontrollers; inter-processor communication through bridge protocols, shared memory and remote procedure calls.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Heterogeneous computing systems
- System-on-chip architecture
- Inter-processor communication
- Shared memory and remote procedure calls
Unit 2: Hardware Abstraction and Power
5 hoursHardware abstraction layers, memory-mapped I/O and peripheral multiplexing; power domains, sleep states, wake-on-interrupt and power-aware scheduling; general-purpose versus task-specific compute units.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Hardware abstraction layers and memory-mapped I/O
- Power domains and sleep states
- Power-aware scheduling
- General-purpose versus task-specific compute
Unit 3: Neural Hardware Acceleration and Vision Pipelines
4 hoursNeural hardware acceleration with SIMD and vectorisation; the vision processing chain: MIPI-CSI, image signal processors and zero-copy frame buffers.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- SIMD and vectorisation
- The vision processing chain: MIPI-CSI and ISPs
- Zero-copy data paths
Unit 4: Specialised Accelerators
5 hoursThe role of DSPs and NPUs in offloading the CPU; computational cost in vision and reducing real-time preprocessing overhead.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- DSPs and NPUs
- Offloading work from the CPU
- Real-time image preprocessing on the edge
Unit 5: Model Compression and Quantisation
5 hoursMoving from FP32 floating point to INT8 fixed point; quantisation principles, scale and zero-point mapping; the effect of lost precision on model accuracy.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Model compression for the edge
- Quantisation: FP32 to INT8
- Scale and zero-point mapping
- Accuracy loss from quantisation
Unit contents are summarised from the official DBATU syllabus . AiCanCode.org is not affiliated with Dr. Babasaheb Ambedkar Technological University (DBATU).