🔍 Bridging the Gap: Amrita ECE/EAC vs Emerging AI Hardware
Analyzing the current curriculum structure at Amrita Vishwa Vidyapeetham and identifying opportunities to incorporate processing-in-memory, neuromorphic computing, photonic computing, and modern AI accelerators.
Amrita Vishwa Vidyapeetham: Current Curriculum Landscape
Based on the 2023 official curricula published on webfiles.amrita.edu[Amrita Src], both B.Tech Electronics and Communication Engineering (ECE) and B.Tech Electronics and Computer Engineering (EAC) programs follow an 8-semester structure with shared first-year foundations.
Shared Foundation
Both curricula share a similar first-year structure with engineering mathematics, programming/problem solving, fundamentals of electrical engineering, physics of semiconductors, labs, technical communication, and courses in Indian heritage/mastery of mind. By semester II both cover Engineering Mathematics II, Network Analysis, labs, and courses like Introduction to Internet of Things.
ECE: Communication, RF, and VLSI Focus
The ECE curriculum emphasizes:
- Signal Processing, Digital Systems, Microprocessors/Microcontrollers
- Electromagnetic Theory, Analog/Digital Circuits, Computer Systems and Architecture
- VLSI Design, Radio Frequency and Microwave Engineering
- Digital Communication, Wireless Communication & Networks
- Computer Networks and Protocols, Embedded Systems
- Professional Electives, Project work, and integrated labs throughout
EAC: Electronics + Computer Systems Blend
The EAC curriculum retains core electronics while adding dedicated computer courses:
- Computer Programming, Computer Hardware and System Essentials
- Object Oriented Programming, Data Structures and Algorithms
- Computer Organization and Architecture, Database Management Systems
- Data Communication and Networks
- Retains VLSI Design, Embedded Systems alongside CS fundamentals
This combination produces a program that sits between ECE and CSE with emphasis on hardware-software integration.
Emerging Research Trends (2025-2026)
1) AI Accelerators
- Shift to specialized accelerators: ASIC/XPU share growing from ~15% in 2024 to ~40% in 2026 in inference[1][6]
- Wafer-scale and dataflow: Cerebras wafer-scale engines deliver massive parallelism; Groq focuses on low-latency, deterministic inference[2][3][5]
- Custom silicon adoption: Google TPUs, AWS Trainium/Inferentia, Microsoft Maia, Broadcom-custom ASICs[1][6][9]
- Market growth: AI accelerators market projected ~$174.69B (2026) → ~$518.12B (2031), CAGR ~24.3%[8]
2) Processing-in-Memory (PIM)
- Memory wall solution: Integrating compute near/inside memory reduces data movement for bandwidth-bound AI workloads[10][12][17][18]
- Commercialization: Samsung HBM-PIM demonstrates acceleration; UPMEM offers commercial DRAM modules with RISC-V PUs[12][13][14][16]
- CXL-PIM emerging: Flexible disaggregated approach gaining traction[10][19]
- Market growth: PIM market projected significant growth (CAGR ~29.4% through ~2034)[15]
3) Neuromorphic Computing
- Spiking neural networks: Event-driven, sparse computation for ultra-low power edge AI[21][24][26][30][32]
- Commercial chips: Intel Loihi/Loihi 2, BrainChip Akida, IBM NorthPole, SpiNNaker 2[22][23][25][27][28][30]
- Scalability demonstrated: Intel's Hala Point (Loihi-based)[21]
- Market growth: Neuromorphic chip market ~$1.84B (2025) → ~$42.84B (2035), CAGR ~34.2%[23]
- Patent surge: ~401% growth in neuromorphic chip patents in 2025[22]
4) Photonic Computing
- Silicon photonics adoption: Optical I/O, co-packaged optics (CPO), and photonic accelerators for bandwidth/energy[33][34][36][39][42][43]
- Commercial players: Lightmatter, Ayar Labs, Celestial AI (acquired by Marvell for $3.25B in Dec 2025)[33][35][37][38][40]
- Data center integration: NVIDIA Spectrum-X/Quantum-X and industry focus on photonics for AI clusters[36][43]
- Market growth: Photonic computing market growing rapidly[34][41]
- Neuromorphic-photonic convergence: Leveraging optics for high-speed, low-latency SNNs[44][45][47][48][49]
Identified Gaps
Based on trends and ABET 2025-2026 criteria[51], several gaps are apparent:
- Lack of non-von Neumann architectures: Most undergrad curricula emphasize CPUs/GPUs; little coverage of PIM, dataflow (Groq/Cerebras), SNNs, or photonic compute[1][10][21][33]
- Memory systems underemphasized: Memory wall, HBM, PIM, CXL are often treated superficially despite being central to AI performance/efficiency[10][12][17]
- Neuromorphic/brain-inspired computing missing: SNNs, event-driven computing, and neuromorphic hardware rarely appear outside specialized grad courses[21][22][25]
- Photonics/computing optics gap: Few ECE programs integrate silicon photonics, optical interconnects, or photonic accelerators into core curriculum despite industry adoption[33][36][39][43]
- Lack of hands-on with emerging accelerators: Students have limited access to PIM (UPMEM), neuromorphic (Loihi/AKIDA dev kits), or photonic simulation tools.
- System-level AI hardware perspective missing: Mapping DNNs to diverse accelerators, co-design, $/token/energy, and heterogeneous systems not well covered[6][9]
- ABET alignment opportunity: ABET 2025-2026 criteria emphasize complex problem solving and modern engineering tools, but programs lag in incorporating these emerging paradigms[51]
Recommendations for Modernization
| Topic | Level | Suggested Content | Hands-on/Tools | Rationale |
|---|---|---|---|---|
| AI Accelerator Fundamentals | Undergrad/Grad | Roofline model, dataflow vs SIMD, systolic arrays, quantization, mapping DNNs to accelerators (TPU/Groq/Cerebras) | Simulators (gem5-Aladdin, SMAUG, Timeloop/Accelergy), case studies of Groq/Cerebras | Core for modern AI systems[6][9] |
| Processing-in-Memory | Grad (with UG elective) | Memory wall, PIM taxonomies (near/in-memory), HBM-PIM, UPMEM, CXL-PIM, tradeoffs | UPMEM SDK (if available), PIM simulators (Ramulator-PIM, PIMSim), coding in PIM model | Critical for bandwidth-bound AI[10][12][14][16] |
| Neuromorphic Computing | Elective/Grad | Spiking neurons, SNN training/inference, event-driven computing, Intel Loihi, BrainChip Akida, applications (sensing/always-on) | Lava (Intel Neuromorphic), Nengo, Brian2, access to Loihi via Intel Neuromorphic Research Cloud | Energy-efficient edge AI[21][25][27][28] |
| Silicon Photonics for AI | Grad/Elective | Waveguides/modulators/detectors, optical interconnects, CPO, photonic accelerators, photonic neuromorphic | Lumerical/PhotonDesign (if licensed), open tools (Meep), Lightmatter/Ayar Labs case studies | Growing in data centers[33][36][39][43] |
| Heterogeneous AI Systems | Capstone/Grad | Co-design, workload mapping, performance/energy modeling, $/token, when to use GPU/ASIC/PIM/neuromorphic/photonic | End-to-end project: map LLM/component to optimal substrate with modeling | Prepares for real system choices[1][4][6] |
Implementation Notes
- Modular integration: Add modules to existing courses (Computer Architecture, VLSI, Optoelectronics) rather than only creating new full courses.
- Industry partnerships: Leverage Intel Neuromorphic Research Cloud, UPMEM educational access, Cerebras/Groq academic programs, and Lightmatter/Ayar Labs collaborations.
- ABET compliance: Tie these to student outcomes (complex problems, modern tools) - document how they address emerging technologies[51].
- Interdisciplinary: Combine ECE with CS (ML systems) and Materials/Photonics to cover full stack.
← Previous: Curriculum Comparison | Home | Next: References →