| Talk | Speaker |
| Title: MINDFULNESS: Mechanisms for Intentionally Normalizing Delays — Frameworks for User Latency to Nudge Engagement Speed Slowly Abstract: Modern computer systems are aggressively optimized for performance. Architectural innovations such as prefetching, speculative execution, hyperthreading, and multicore parallelism have significantly reduced application latency and increased throughput. While these advances improve system efficiency, they have also reshaped patterns of user interaction: applications respond instantly, media becomes shorter and more fragmented, and users increasingly expect immediate gratification. As a result, extreme responsiveness unintentionally reinforces distraction and fragmented attention. This work introduces MINDFULNESS, an architectural framework that intentionally introduces controlled delays into computing systems to encourage slower and more deliberate user engagement. MINDFULNESS detects behavioral indicators of distraction --- such as rapid application switching or bursty process creation --- and dynamically activates architectural interventions designed to promote focus. These interventions deliberately restrict or disable performance optimizations such as speculation and hyperthreading. MINDFULNESS raises a broader question for computer architecture: should systems always optimize exclusively for speed, or should they also optimize for human well-being? | Deepanjali Mishra (Carnegie Mellon University), Ramya Prabhu (École Polytechnique Fédérale de Lausanne), Aditya K Kamath (University of Washington) |
| Title: Is Consciousness the Brain's Consistency Model? Abstract: Consciousness is one of the oldest and most important open problems in science. It is now also one of the most urgent given we are building multi-agent AI systems. Whether we want to engineer consciousness into such systems or ensure we do not build it accidentally, we need the same thing: a precise understanding of the necessary conditions for consciousness. Our thesis is that the tools computer system architects developed to tame the concurrency of multiprocessors—specifically consistency models and litmus tests—are precisely the right tools for this problem. But what is the connection between the concurrency of a multiprocessor and consciousness? The brain is a massively parallel system: distinct neural subsystems—visual, auditory, olfactory, among others—process their inputs simultaneously. Yet we experience a single, serial stream of consciousness. Global Workspace Theory (GWT), a leading theory of consciousness, explains why: these parallel processors compete for access to a limited-capacity workspace; one wins; the winning content is broadcast globally to all processors. The serial sequence of winners is the stream of consciousness. To a computer system architect, this is immediately recognizable: GWT describes a consensus protocol enforcing a serializable history. | Vijay Nagarajan (University of Utah) |
| Title: Space-Based RF Solar Power: Wireless Energy for Edge Devices Abstract: Space-based solar power is a transformative approach to renewable energy generation, wherein solar power harvested by satellites in orbit is transmitted to Earth via radio frequency (RF) beams. We show the evolving history of satellite-based wireless power transfer (WPT), from the motivation for space-based energy harvesting and foundational technologies in microwave power transmission, to recent advances in beamforming and intelligent reflecting surfaces, and current critical technical challenges that must be addressed for practical deployment. We show that while significant progress has been made in phased array beamforming, rectenna efficiency, and system integration, challenges remain in beam alignment, atmospheric attenuation, and regulatory frameworks. A successful space-based RF power delivery can make a significant impact, possibly reshaping current understanding of computer architecture, leading to a cloud-free, power-unconstrained computing framework. | Yiwen Song, Swarun Kumar (Carnegie Mellon University) |
| Title: The Asemantic Code Revolution Abstract: Every abstraction layer in the modern software stack exists to solve legitimate problems—coordinating independent developers and enforcing trust across boundaries. However, each layer exacts a Cognitive Tax: overhead paid not for correctness, but for human coordination. Published measurements bound this non-computational overhead at ∼59% across the ISA frontend, ABI, IEEE 754 logic, and per-die guard-band slack. We propose a shift from shipping static artifacts to distributing formal intent. Developer intent is translated into Z3 invariant bundles, which a local NeuralSymbolic Oracle then synthesizes into ‘Asemantic Code’—an artifact governed by load-time proof certificates and mathematically mutated to exploit the specific manufacturing physics of its execution die | Puneetha Ramachandra (Google) |
| Title: SICC: Server Integrated Carbon Capture Abstract: The climate science is clear: we need to curb our carbon emissions. Unfortunately, data centers are currently part of the problem. What if, instead of increasing global carbon emissions, data centers could actively reduce net emissions? We propose Server Integrated Carbon Capture (SICC), which enables data centers to capture carbon on-site. SICC adds direct air capture to data centers and reuses surplus heat from servers to supply heat needed for direct air capture. Adding SICC to data centers has implications for the data center hardware-software stack, including job scheduling, power budgets, and server component selection. To maximize the benefits of SICC, we should design SICC-aware systems. | Abnash Bassi (Carnegie Mellon University) |
Pittsburgh, USA – March 22-26, 2026.