Thermal and IR Drop Analysis Using Convolutional Encoder-Decoder Networks

Chhabria, Vidya A.; Ahuja, Vipul; Prabhu, Ashwath; Patil, Nanasaheb M; Jain, Pooja; Sapatnekar, Sachin S.

doi:10.1145/3394885.3431583

Cited by 30 publications

(6 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This work represents the development and maturation of a previous conference publication [7] that addressed the static and transient thermal analysis problems and static IR drop analysis problem using encoder decoder based generative (EDGe) networks. We extend this to demonstrate the use of EDGe networks for transient IR drop analysis and the EM hotspot classification.…”

Section: Overview Of Our Approachmentioning

confidence: 99%

Encoder-Decoder Networks for Analyzing Thermal and Power Delivery Networks

Chhabria¹,

Ahuja²,

Prabhu³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Power delivery network (PDN) analysis and thermal analysis are computationally expensive tasks that are essential for successful IC design. Algorithmically, both these analyses have similar computational structure and complexity as they involve the solution to a partial differential equation of the same form. This paper converts these analyses into image-to-image and sequence-to-sequence translation tasks, which allows leveraging a class of machine learning models with an encoder-decoder-based generative (EDGe) architecture to address the time-intensive nature of these tasks. For PDN analysis, we propose two networks: (i) IREDGe: a full-chip static and dynamic IR drop predictor and (ii) EMEDGe: electromigration (EM) hotspot classifier based on input power, power grid distribution, and power pad distribution patterns. For thermal analysis, we propose ThermEDGe, a full-chip static and dynamic temperature estimator based on input power distribution patterns for thermal analysis. These networks are transferable across designs synthesized within the same technology and packing solution. The networks predict on-chip IR drop, EM hotspot locations, and temperature in milliseconds with negligibly small errors against commercial tools requiring several hours.

show abstract

Section: Overview Of Our Approachmentioning

confidence: 99%

Encoder-Decoder Networks for Analyzing Thermal and Power Delivery Networks

Chhabria¹,

Ahuja²,

Prabhu³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Recently, with the popularity of machine learning, a few works deployed various machine learning techniques to accelerate the computation of supply noise or electro-migration prediction without actually invoking the sparse linear system simulator [10][11][12][13][14][15][16][17]. In order to achieve accurate prediction, the detailed PDN structural and electrical information are extracted as input features to train the model, such as instance power and path resistance [12][13][14][15].…”

Section: Introductionmentioning

confidence: 99%

“…However, such instance-level information and path resistance itself demands power or static power delivery analysis, which actually involves implicit but non-trivial training overhead. In addition, while most existing works placed focus on static IR drop of PDN [10][11][12], dynamic (or transient) PDN noise is actually more important for sign-off, which is triggered by the resonance between package and die and hence results in more severe noise that needs to be validated. Finally, since the size of a commercial PDN can be tremendous (millions to billions of nodes), the direct employment of deep learning techniques to predict node voltage may easily result in a too huge neural network with serious scalability issue [11].…”

Section: Introductionmentioning

confidence: 99%

“…In addition, while most existing works placed focus on static IR drop of PDN [10][11][12], dynamic (or transient) PDN noise is actually more important for sign-off, which is triggered by the resonance between package and die and hence results in more severe noise that needs to be validated. Finally, since the size of a commercial PDN can be tremendous (millions to billions of nodes), the direct employment of deep learning techniques to predict node voltage may easily result in a too huge neural network with serious scalability issue [11]. Thus, many works have to either simplify the input to the neural network or check the noise region by region [13][14][15], inevitably compromising accuracy or efficiency.…”

Section: Introductionmentioning

confidence: 99%

“…When dealing with tens of test vectors for dynamic analysis, it is very time-consuming to complete all the vectors by deploying the conventional simulation based methods [5][6][7][8][9]. Recently, various machine learning based algorithms have been proposed for static IR drop analysis [10][11][12]. For example, the work in [12] proposed an incremental IR drop prediction model, which feeds the structural and electrical features of PDN into XGBoost to estimate the static IR drop.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Worst-Case Dynamic Power Distribution Network Noise Prediction Using Convolutional Neural Network

Xiao¹,

Chen²,

Yin³

et al. 2022

Preprint

View full text Add to dashboard Cite

Worst-case dynamic PDN noise analysis is an essential step in PDN sign-off to ensure the performance and reliability of chips. However, with the growing PDN size and increasing scenarios to be validated, it becomes very time-and resource-consuming to conduct full-stack PDN simulation to check the worst-case noise for different test vectors. Recently, various works have proposed machine learning based methods for supply noise prediction, many of which still suffer from large training overhead, inefficiency, or non-scalability. Thus, this paper proposed an efficient and scalable framework for the worst-case dynamic PDN noise prediction. The framework first reduces the spatial and temporal redundancy in the PDN and input current vector, and then employs efficient feature extraction as well as a novel convolutional neural network architecture to predict the worst-case dynamic PDN noise. Experimental results show that the proposed framework consistently outperforms the commercial tool and the state-of-the-art machine learning method with only 0.63-1.02% mean relative error and 25-69× speedup.

show abstract

Deep Learning for Analyzing Power Delivery Networks and Thermal Networks

Chhabria

Sapatnekar

2022

Machine Learning Applications in Electronic Design Automation

View full text Add to dashboard Cite

Thermal and IR Drop Analysis Using Convolutional Encoder-Decoder Networks

Cited by 30 publications

References 5 publications

Encoder-Decoder Networks for Analyzing Thermal and Power Delivery Networks

Encoder-Decoder Networks for Analyzing Thermal and Power Delivery Networks

Worst-Case Dynamic Power Distribution Network Noise Prediction Using Convolutional Neural Network

Deep Learning for Analyzing Power Delivery Networks and Thermal Networks

Contact Info

Product

Resources

About