/Power Delivery Challenges In Angstrom Nodes: Paving the Way For 3D Integrated Circuits (3D-ICs) for high power density AI/ML applications

Power Delivery Challenges In Angstrom Nodes: Paving the Way For 3D Integrated Circuits (3D-ICs) for high power density AI/ML applications

PhD - Leuven | Just now

Ready to embark on a journey towards making future compute systems energy efficient? Join us for this exciting opportunity to tackle power delivery challenges in heterogenous 3DICs to enable technologies that have superior performance, are reliable, and energy-efficient for AI-driven compute demands!!

Semiconductor technology driven by Moore’s law has reached an inflection point due to physical and technological barriers. Further innovation is fueled by the disintegration of various functional components in an IC and subsequent re-integration using advanced packaging techniques like 2.5D and 3D system integration. The separate and more tailored optimization and integration, also popularly known as heterogeneous integration (HI) allows for achieving higher performance per Watt of power. In this context, IMEC’s CMOS 2.0 paradigm[1] views separate tiers for high drive logic, high-density logic, level 1 (L1) and level 2 (L2) cache memories (SRAMs), and power delivery network (PDN), clock delivery network (CDN), and I/Os.

However, several challenges come in the way of this attractive option: among which power delivery and thermal challenges are the foremost. Since stacking of active layers inherently implies higher power density, power supply network impedance can become prohibitively large to keep the system functional without errors. The rising power density (due to higher current per footprint) and faster transients (due to faster switching speeds) lead to higher IR and LdI/dt noise injected into the power lines respectively. With pitch scaling, the interconnect resistance has gone up significantly leading to even larger IR drops. Since the power supply impedance (considering package and board RLC) hasn’t really kept up with the scaling, the supply noise issue is critical, especially in 3D designs where the inter-layer vias (ILVs) or through Silicon vias (TSVs) add to more parasitics. The frequency dependence of power supply impedance makes the impact sensitive to workloads.     

This PhD opportunity offers the opportunity to take up these exciting  challenges and to design optimization methodologies, solutions and trade-off analysis for power delivery and management in heterogenous 3D-ICs for high peak load AI/ML applications  e.g. design technology optimization methodologies using imec’s advanced passive component technology like high-density DCAPs, enabling fully integrated voltage regulators (FIVR). You will work with IMEC’s advanced technology PDK suited for Angstrom nodes to implement commercial design benchmarks executing realistic workloads. You will be using electromagnetic field simulator (e.g., High-Frequency Structure Simulator (HFSS) from ANSYS) to accurately model package and board level as well as TSV parasitics for different flavours of 2.5D and integration technologies, e.g., F2F/F2B hybrid bonding Effect of running workloads will be studied to identify noise-critical windows. Finally, you will explore opportunities to mitigate the supply noise by strategically placing decoupling capacitors/ DCAPs (e.g. on-chip 2.5D MIMCAP). Recently, backside power delivery network (BSPDN) has been introduced to improve power integrity[2] of the wafer for more generous routing resource allocation for signals in the front side, thereby making the power lines fatter and less resistive. The backside of the wafer can be further exploited to include power switches and /or incorporating high-density MIMCAPs to filter high-frequency noise and for realizing power-efficient designs. for realizing power-efficient designs.

More specifically, your work will involve the following activities:

  • Implementation of commercial benchmark digital designs with Cadence tools
  • Power analysis of complex 3D-IC physical implementations
  • Power delivery network modeling and simulation
  • Design of optimization methodologies for
    • Supply noise mitigation
    •  Power, Performance, Area (PPA) penalty (e.g., leakage increase due to DCAP insertion)
    • Increase the functionality of the wafer’s backside with different power delivery modules and evaluate PPA metrics thereof


Required background: Electrical/electronics engineering

Type of work: 10% circuit design and implementation, 20% design automation and coding, 30% physical design, 30% 3D interconnect component modeling/simulation, 10% literature

Supervisor: Bertrand Parvais

Daily advisor: Priya Venugopal

The reference code for this position is 2026-099. Mention this reference code on your application form.

Who we are
Accept analytics-cookies to view this content.
imec's cleanroom
Accept analytics-cookies to view this content.

Send this job to your email