Seah Kim

Seah Kim

Auto-generated profile, based on publicly available information

Seah Kim

University of California, Berkeley, Berkeley, CA, USA

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 March 2023
Verified
Conference Paper
Open access
SKSeah Kim
Seah Kim
HKHyoukjun Kwon
Hyoukjun Kwon
JSJinook Song
Jinook Song
JJJihyuck Jo
Jihyuck Jo
YCYu-Hsin Chen
Yu-Hsin Chen
VCVikas Chandra
Vikas Chandra
Emerging real-time multi-model ML (RTMM) workloads such as AR/VR and drone control involve dynamic behaviors in various granularity; task, model, and layers within a model. Such dynamic behaviors introduce new challenges to the system software in an ML system since the overall system load is not completely predictable, unlike traditional ML workloads. In addition, RTMM workloads require real-time processing, involve highly heterogeneous models, and target resource-constrained devices. Under such circumstances, developing an effective scheduler gains more importance to better utilize underlying hardware considering the unique characteristics of RTMM workloads. Therefore, we propose a new scheduler, DREAM, which effectively handles various dynamicity in RTMM workloads targeting multi-accelerator systems. DREAM quantifies the unique requirements for RTMM workloads and utilizes the quantified scores to drive scheduling decisions, considering the current system load and other inference jobs on different models and input frames. DREAM utilizes tunable parameters that provide fast and effective adaptivity to dynamic workload changes. In our evaluation of five scenarios of RTMM workload, DREAM reduces the overall UXCosT, which is an equivalent metric of the energy-delay product (EDP) for RTMM defined in the paper, by 32.2% and 50.0% in the geometric mean (up to 80.8% and 97.6%) compared to state-of-the-art baselines, which shows the efficacy of our scheduling methodology.


30 March 2025
Verified
Conference Paper
Open access
SKSeah Kim
Seah Kim
RHRoger Hsiao
Roger Hsiao
BNBorivoje Nikolic
Borivoje Nikolic
JDJames Demmel
James Demmel
YSYakun Sophia Shao
Yakun Sophia Shao
Simultaneous Localization and Mapping (SLAM) plays a crucial role in robotics, autonomous systems, and augmented and virtual reality (AR/VR) applications by enabling devices to understand and map unknown environments. However, deploying SLAM in AR/VR applications poses significant challenges, including the demand for high accuracy, real-time processing, and efficient resource utilization, especially on compact and lightweight devices. To address these challenges, we propose SuperNoVA, which enables high-accuracy, real-time, large-scale SLAM in resource-constrained settings through a full-stack system, spanning from algorithm to hardware. In particular, SuperNoVA dynamically constructs a subgraph to meet the latency target while preserving accuracy, virtualizes hardware resources for efficient graph processing, and implements a novel hardware architecture to accelerate the SLAM backend efficiently. Evaluation results demonstrate that, for a large-scale AR dataset, SuperNoVA reduces full SLAM backend computation latency by 89.5% compared to the baseline out-of-order CPU and 78.6% compared to the baseline embedded GPU, and reduces the maximum pose error by 89% over existing SLAM solutions, while always meeting the latency target.

28 October 2023
Verified
Conference Paper
Open access
SKSeah Kim
Seah Kim
JZJerry Zhao
Jerry Zhao
KAKrste Asanovic
Krste Asanovic
BNBorivoje Nikolic
Borivoje Nikolic
YSYakun Sophia Shao
Yakun Sophia Shao
With the widespread adoption of deep neural networks (DNNs) across applications, there is a growing demand for DNN deployment solutions that can seamlessly support multi-tenant execution. This involves simultaneously running multiple DNN workloads on heterogeneous architectures with domain-specific accelerators. However, existing accelerator interfaces directly bind the accelerator’s physical resources to user threads, without an efficient mechanism to adaptively re-partition available resources. This leads to high programming complexities and performance overheads due to sub-optimal resource allocation, making scalable many-accelerator deployment impractical. To address this challenge, we propose AuRORA, a novel accelerator integration methodology that enables scalable accelerator deployment for multi-tenant workloads. In particular, AuRORA supports virtualized accelerator orchestration via co-designing the hardware-software stack of accelerators to allow adaptively binding current workloads onto available accelerators. We demonstrate that AuRORA achieves 2.02 higher overall SLA satisfaction, 1.33 overall system throughput, and 1.34 overall fairness compared to existing accelerator integration solutions with less than 2.7% area overhead.

13 June 2024 in IEEE Micro
Verified
Journal Article
Closed access
SKSeah Kim
Seah Kim
JZJerry Zhao
Jerry Zhao
KAKrste Asanović
Krste Asanović
BNBorivoje Nikolić
Borivoje Nikolić
YSYakun Sophia Shao
Yakun Sophia Shao
To meet the increasingly demanding compute requirements of modern workloads, SoCs must provide an accelerator-rich hardware architecture and software programming interface. However, scalability remains a first-order concern, as introducing additional unmanageable complexity to either physical design or software integration may prohibit the deployment of new accelerators. To address these challenges, this work presents AuRORA, an accelerator integration methodology that provides a scalable physical accelerator interface, while preserving software semantics with minimal overhead for accelerator access. AuRORA provides a new accelerator integration methodology that preserves the software and hardware interface of a tightly-CPU-coupled accelerator, while physically disaggregating the accelerators away from a host CPU. To address software scalability, AuRORA also includes a lightweight software runtime for an SoC with heterogeneous accelerators, providing low-overhead access to these accelerators for multi-tenant applications.

21 January 2026 in IEEE Journal of Solid-State Circuits
Verified
Journal Article
Closed access
SKSeah Kim
Seah Kim
JZJerry Zhao
Jerry Zhao
RHRoger Hsiao
Roger Hsiao
YCYufeng Chi
Yufeng Chi
VIVighnesh Iyer
Vighnesh Iyer
YSYakun Sophia Shao
Yakun Sophia Shao
Three-dimensional reconstruction in robotics combines heterogeneous workloads: INT8 DNNs for depth estimation (DE) and feature extraction (FE), and FP32 sparse linear algebra (LA) workloads for simultaneous localization and mapping (SLAM). These tasks stress heterogeneous computing, accelerator integration, and scheduling. We present MAVERIC, a 16-mm2 system-on-chip (SoC) in Intel 16 with 4 CPU cores and 13 accelerators (eight INT8, five FP32) that executes diverse workloads, including the full DE and SLAM pipeline end-to-end, including loop closure (LC). A hierarchical, generator-driven SoC flow and a multi-layer 2-D-mesh network-on-chip (NoC) streamline integration of on-chip components and scale the design. A lightweight remote CPU–accelerator interface with a custom hardware protocol enables efficient scheduling and parallel utilization of heterogeneous accelerators. We compress the vocabulary tree (VT) for on-chip LC and map numerical optimization tasks to FP32 accelerators, providing full on-chip SLAM support. MAVERIC operates at up to 1 GHz, achieves 8 TOPS/W peak efficiency, and processes 72 FPS 10 mJ/frame for end-to-end DE and SLAM. These results demonstrate that mixed-precision, many-accelerator SoCs are an effective architecture for real-time, energy-efficient robotics perception.