Special Topic on 3D Heterogeneous Integration for Energy-Efficient AI Systems

Amit Ranjan Trivedi, University of Illinois Chicago,

Jaydeep Kulkarni, University of Texas Austin,

Asif Khan, Georgia Institute of Technology,

The energy and performance of contemporary artificial intelligence systems are increasingly constrained by data movement, memory access, and interconnect bandwidth rather than by arithmetic throughput alone. While continued CMOS scaling has enabled steady improvements in compute density, it offers limited relief for these system-level bottlenecks. Three-dimensional heterogeneous integration (3DHI) provides a complementary approach by enabling vertical integration of logic, memory, sensors, and specialized accelerators within a single system.

3DHI encompasses a range of integration techniques, including through-silicon vias, hybrid bonding, monolithic 3D integration, interposers, and chiplet-based assemblies. These techniques enable tighter coupling between system components, reduced interconnect energy, and new architectural organizations that are difficult to realize in planar designs. When applied to AI workloads, 3DHI enables alternative system partitions, memory hierarchies, and sensor-compute interfaces that can improve energy efficiency, latency, and bandwidth utilization.

This Special Topic focuses on AI systems and architectures enabled by 3D heterogeneous integration, with emphasis on cross-layer co-design spanning devices, circuits, architectures, and algorithms. Contributions should clearly articulate how 3D integration influences system behavior, performance, and energy-efficiency and where possible provide quantitative comparisons to planar or conventional baselines. They should also address how the thermal challenges posed by excessive power densities is going to be mitigated (and how well) in any proposed 3D Integration approach. Both experimental demonstrations and well-supported modeling or simulation studies are within scope.

Topics of interest include, but are not limited to:

1. Applications: Relevant application domains include, but are not limited to: Edge, embedded, and embodied AI; Autonomous and robotic systems; Data-center and cloud AI accelerators; Scientific and sensor-driven AI workloads
2. 3D Integrated AI Architectures: Logic-on-memory and memory-on-logic AI accelerators; 3D-stacked SRAM, DRAM, and non-volatile memory for AI workloads; Monolithic 3D versus TSV-based AI system implementations; Vertical interconnects; on-stack networks; and bandwidth-energy tradeoffs
3. Near-Sensor and In-Sensor AI Systems: 3D integration of image, LiDAR, radar, and event-based sensors with compute, Sensor–memory–logic co-integration for low-latency inference, System-level evaluation of energy and data-movement reductions
4. Heterogeneous Integration with Emerging Devices: Integration of analog or mixed-signal compute blocks within 3D AI systems; CMOS integration with resistive, ferroelectric, photonic, or spintronic elements; Reliability, thermal, and variability considerations in heterogeneous stacks
5. Algorithm and Architecture Co-Design: AI models and mappings optimized for vertically integrated memory hierarchies; Partitioning of neural networks across tiers in 3D systems; Training and inference strategies that account for 3D communication constraints
6. Modeling, Design Automation, and Evaluation: Cross-layer modeling frameworks for 3DHI AI systems; Thermal-aware design and performance analysis; Energy, latency, and area benchmarking methodologies
7. Prototype Demonstrations and Case Studies: Silicon or system-level prototypes of 3D-integrated AI systems; Measured or simulated comparisons with planar implementations; Application-driven evaluations in edge, embedded, or data-center contexts
8. System-Level Challenges: Power delivery and thermal management in 3D-integrated AI systems; Yield, test, and scalability considerations; Chiplet-based AI systems and heterogeneous integration strategies

Submit your paper through the JXCDC submission site.

• Open for Submission: February 15, 2026
• Submission Deadline: May 15, 2026
• First Notification: June 15, 2026
• Revision Submission: July 1, 2026
• Final Decision: July 15, 2026
• Online Special Topic Publication: August 1, 2026

Papers submitted earlier than the submission deadline will be reviewed upon submission and if accepted will get published earlier than the timeline listed above.