<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[logic-synthesis-in-vlsi - Medium]]></title>
        <description><![CDATA[This blog is about logic synthesis in vlsi, synthesis flow and machine learning functionality in logic synthesis - Medium]]></description>
        <link>https://medium.com/logic-synthesis-in-vlsi?source=rss----84dd639248b1---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>logic-synthesis-in-vlsi - Medium</title>
            <link>https://medium.com/logic-synthesis-in-vlsi?source=rss----84dd639248b1---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 09:48:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/logic-synthesis-in-vlsi" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[High Level Synthesis using Linear Regression]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/high-level-synthesis-using-linear-regression-a2ae6676f1d7?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/a2ae6676f1d7</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[vlsi]]></category>
            <category><![CDATA[high-level-synthesis]]></category>
            <category><![CDATA[linear-regression]]></category>
            <dc:creator><![CDATA[Harshada Belgi]]></dc:creator>
            <pubDate>Fri, 30 Apr 2021 01:21:35 GMT</pubDate>
            <atom:updated>2021-04-30T01:21:34.980Z</atom:updated>
            <content:encoded><![CDATA[<p>High Level Synthesis (HLS) works at the algorithmic level in hardware design systems. HLS tools provide automatic conversion from C/C++/SystemC-based specification to hardware description languages like Verilog or VHDL. It has been inferred that these HLS tools are improving the productivity in customised hardware design systems. In certain cases of HLS, the long synthesis time for each design brings about a restriction on design space exploration (DSE). DSE is the process of finding some design solution which optimally meets the design requirements from a space of tentative design points. This exploration is quite complex considering the various levels of abstraction and the entire process from selection of parameter values to choosing an algorithm and then optimising it.</p><p><strong>Machine Learning for High Level Synthesis:</strong></p><p>Machine Learning techniques are applied to improve the performance of HLS tools with 3 major advantages:</p><p>1. Fast and accurate result estimation</p><p>2. Refining the conventional DSE Algorithms</p><p>3. Reforming DSE as active learning problem</p><p><strong>Performance Prediction using ML:</strong></p><p>While HLS-generated register-transfer-level (RTL) models are in general not human-readable, HLS tools provide a set of reports to quantitatively convey the expected performance, timing, resource usage, and composition of the synthesized RTL design. ML can be used to improve the accuracy of HLS reports through learning from real design benchmarks. Multiple supervised learning techniques like Linear Regression, Random Forest, Support Vector Machine, etc. are used to test the quality of the result through data processing training estimation models. Features can be extracted from the HLS reports to accurately predict post-implementation results without actually running the implementation flow. Using these features and training the ML models, we can achieve high accuracy for the estimation tasks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/429/1*zkf6cdE5vNC_e_uFZswJNA.png" /><figcaption>HLS Tool Flow with Machine Learning Model</figcaption></figure><p><strong>Linear Regression for Performance Prediction:</strong></p><p>Linear regression is a prediction oriented Machine Learning algorithm which attempts to model the relationship between two variables by fitting a linear equation to observed data.</p><p>To enable ML technique for HLS estimation, a dataset is built up of implementation results. Features are extracted from this dataset, redundant features are then removed and irrelevant features are eliminated. Regression models are trained to estimate post-implementation resource usages as well as classification models to predict whether the target clock period is met for the implemented design.</p><p>The classic linear regression model, ybi = xi &gt;w, models the target ybi as a linear combination of features xi. Linear regression fits this model onto the training data to determine the w such that a loss function is minimized, where w represents the vector of coefficients for the learned model. We use the Lasso linear model with a loss function of Xw − y k 2 2 + γ kwk1 to train a linear model that minimizes the least-square penalty on the training data with L1 regularization. By tuning the hyperparameter γ, the L1 regularization term γ kwk1 allows us to induce various degree of sparsity into w and in turn regulate the complexity of the model.</p><p>The accuracy of the model can be calculated using the relative absolute error (RAE).</p><p><strong>References:</strong></p><ol><li>Machine Learning for Electronic Design Automation: A Survey</li><li>Pyramid: Machine Learning Framework to Estimate the Optimal Timing and Resource Usage of a High-Level Synthesis Design</li><li>Fast and Accurate Estimation of Quality of Results in High-Level Synthesis with Machine Learning</li><li><a href="https://www.sciencedirect.com/topics/computer-science/design-space-exploration">https://www.sciencedirect.com/topics/computer-science/design-space-exploration</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a2ae6676f1d7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/high-level-synthesis-using-linear-regression-a2ae6676f1d7">High Level Synthesis using Linear Regression</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Logic synthesis using CNN]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/logic-synthesis-using-cnn-afa1f19f0fe7?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/afa1f19f0fe7</guid>
            <dc:creator><![CDATA[PRANAV BAPAT]]></dc:creator>
            <pubDate>Thu, 29 Apr 2021 17:33:03 GMT</pubDate>
            <atom:updated>2021-04-29T17:33:03.424Z</atom:updated>
            <content:encoded><![CDATA[<p>Introduction</p><p>Logic Synthesis is used in order to realize RTL(Register transfer Level) design which is obtained via a hardware description language(HDL). Various types of machine learning techniques are used in order to optimize the complete process. Convolutional Neural Network(CNN) is one of the many methods used and a preliminary one for this purpose. It can be used from logic synthesis, in placement and routing as well as in power deliver network synthesis and IR drop predictions. CNN models are composed of convolutional layers and other basic blocks such as non-linear activation functions and down sample pooling functions. In the logic synthesis and physical design stage, there are many key sub-problems that can benefit from the power of ML models. Let us see how CNN can be implemented in the above mentioned parts for useful optimizations.</p><p>Logic synthesis:</p><p>This is an optimization problem with complicated constraints, which requires accurate solutions. Hence, using ML algorithms to directly generate logic synthesis solutions is a bit difficult. There are many logic transformations in current synthesis tools. To select an appropriate synthesis flow we need to formulate a multi-class classification problem and design a CNN to map a synthesis flow. The prediction on unlabeled flows are then used to select the optimal synthesis flow. The CNN takes the one-hot encoding of synthesis flows as inputs and outputs the possibilities of the input flow. Sometimes, reinforcement learning is also employed for logic synthesis for better optimization.</p><p>Placement &amp; Routing:</p><p>The basic requirements of routing design rules must be considered in the placement stage. However, it is difficult to predict routing information in the placement stage accurately and fast, hence, recently machine learning is employed to solve this. It was first used for design rule checking (DRC) hotspot detection. The input features of a customized fully convolutional network (FCN) include the outputs of rectangular uniform wire density. It can also be employed to predict design rule violation count. Various other advanced networks are also used in this case, for better optimization and prediction.</p><p>Power deliver network synthesis:</p><p>Power delivery network (PDN) design is a complex iterative optimization task, which strongly influences the performance, area and cost of a chip. To reduce the design time, ML-based algorithms are being recently used. The main challenge is increasing complexity of chip design. The IR drop can be divided into two parts: static and dynamic. A CNN can be used to predict the IR values of different locations and models IR drop estimation problem as a regression task. Further, CNN is used to choose the template for floorplan and placement stages.</p><p>Conclusion:</p><p>Thus from this we can conclude that with the help of various machine learning techniques like CNN we can get better and optimized results in various steps of logic synthesis, by spending less time, however with higher computational requirements.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=afa1f19f0fe7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/logic-synthesis-using-cnn-afa1f19f0fe7">Logic synthesis using CNN</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Lithography Hotspot Detection Using Support Vector Machine]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/lithography-hostpot-detection-using-svm-63ea2f1ef723?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/63ea2f1ef723</guid>
            <category><![CDATA[support-vector-machine]]></category>
            <category><![CDATA[lithographies]]></category>
            <category><![CDATA[supervised-learning]]></category>
            <category><![CDATA[hotspot-detection]]></category>
            <dc:creator><![CDATA[Anmol Salvi]]></dc:creator>
            <pubDate>Thu, 29 Apr 2021 16:07:53 GMT</pubDate>
            <atom:updated>2021-04-29T16:08:55.944Z</atom:updated>
            <content:encoded><![CDATA[<p>INTRODUCTION: Lithography can be defined as the process of transferring patterns of a particular geometric shape in a mask to a thin layer of radiation sensitive material called the resist , thus covering the surface of a semiconductor wafer . Lithography hotspots are places which are susceptible to having fatal pinching (open circuits) or bridging (short circuit) , which occurs due to poor printability of some patterns in a given design layout. In order to avoid such undesirable patterns in a layout , it is mandatory to find hotspots in the early design phase. As manufacturing conditions evolve , lithography hotspot detection faces many challenges which include:-</p><p>a)Real hotspots are extremely hard to find during early design stages and also hard to fix in the post layout phase.</p><p>b)False alarms are detected which result in expensive post processing hotspot removal.</p><p>c)Full chip physical verification as well as optimization requires an extremely fast turn-around time.</p><p>The above problems can be solved by using an appropriate ML algorithm to save time and provide accurate predictions.</p><p>IMPLEMENTATION: SVM which stands for Support Vector Machines , can be used in order to develop a faster and more accurate hotspot identifier and in turn a better model for lithography hotspot detection . It is a supervised learning algorithm and a multiclass classifer. The first step involves training , which is carried out by providing the model with the required values from the dataset , where these values are first normalized to a scale ranging between [-1 , 1]. The model will assign weights to different parameters and return support vectors. More the data it is trained with , more is the error reduced by constantly updating the weights and gradients iteratively. Training targets are also chosen and accurate labels are provided to the targets.When the SVM model is fully trained and configured, we can apply it to evaluate a new design pattern without using accurate lithography simulations.</p><p>ADVANTAGES: a)Faster than already available methods.</p><p>b)More accurate and efficient.</p><p>c)The problem of detecting false alarms is reduced considerably.</p><p>CONCLUSION: Thus one can conclude from the above points just how essential it is to employ machine learning techniques such as SVM in order to develop a highly powerful , fast and efficient model.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=63ea2f1ef723" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/lithography-hostpot-detection-using-svm-63ea2f1ef723">Lithography Hotspot Detection Using Support Vector Machine</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[High Level Synthesis using Random Forest Algorithm]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/high-level-synthesis-using-random-forest-algorithm-9a1cea9ca7a4?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/9a1cea9ca7a4</guid>
            <category><![CDATA[supervised-learning]]></category>
            <category><![CDATA[random-forest]]></category>
            <category><![CDATA[high-level-synthesis]]></category>
            <dc:creator><![CDATA[Anmol Salvi]]></dc:creator>
            <pubDate>Thu, 29 Apr 2021 15:52:38 GMT</pubDate>
            <atom:updated>2021-04-29T16:04:40.659Z</atom:updated>
            <content:encoded><![CDATA[<p>High Level Synthesis using Random Forest Algorithm</p><p>INTRODUCTION:-High level Synthesis is a method by which one can realize RTL(Register transfer Level) by taking an algorithmic description as input. The control algorithms are typically written in high level programming languages such as C , OpenCL and SystemC among others. The adoption of HLS however at an industrial level is still at the evaluation stage. There are 2 major bottlenecks that arise from the use of HLS tools:-</p><p>1)DSE(Design Space Exploration)-A substantial effort is needed for setting the micro-architecture constraints.</p><p>2)The long runtime of such HLS tools.</p><p>IMPLEMENTATION:-Before discussing the implementation its is crucial to know how the random forest algorithm works. The Random Forest Algorithm put simply is a supervised learning algorithm that builds multiple decision trees and merges them together to get a more accurate and stable prediction. Moving on to the implementation , two stages exist , the first being the training stage , followed by the refinement stage . In the training stage , knob settings are sampled in order to get our training set. Post this once our model begins to predict HLS results for all knob settings , the predicted Pareto set values are used to train the model with this Pareto set’s knob settings , which results in updating the training examples , retraining and updating parameters till a highly efficient learning model is achieved.</p><p>ADVANTAGES:-a)HLS knobs which provide binary choices are common. Tree based RF models can very easily handle such binary decisions by introducing a node with 2 branches separating the two decisions.</p><p>b)The RF algorithm is essentially composed of multiple regression trees . When a training set is provided it is internally and randomly partitioned to train individual trees.The final decision is then made by the collective vote from these individual trees. These two steps combined allow the reduction of generalization errors and prediction variance.</p><p>CONCLUSION:-Thus from the above points one can conclude that the way to go forward with the adoption of HLS in industries is to ensure fast and accurate results , which can be provided by integrating a supervised machine learning algorithm such as Random Forest with our HLS tools.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a1cea9ca7a4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/high-level-synthesis-using-random-forest-algorithm-9a1cea9ca7a4">High Level Synthesis using Random Forest Algorithm</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mask Optimization using GAN]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/mask-optimization-using-gan-3c82dcad6881?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/3c82dcad6881</guid>
            <dc:creator><![CDATA[ISHAN APHALE]]></dc:creator>
            <pubDate>Thu, 29 Apr 2021 13:42:50 GMT</pubDate>
            <atom:updated>2021-04-29T13:50:23.192Z</atom:updated>
            <content:encoded><![CDATA[<p>Lithography is one of the key steps in semiconductor manufacturing, which turns the designed circuit and layout into real objects. Two popular research directions are lithography hotspot detection and types of mask optimizations. We will focus on mask optimization in this blog. Mask optimization tries to compensate diffraction information loss of design patterns such that the remaining pattern after lithography is very close to the design patterns. Mask optimization plays an important role in VLSI design. Optical proximity correction (OPC) and sub-resolution assist feature (SRAF) insertion are two main methods to increase the printability of the target pattern, and we will be looking into them with the point of view of GANs. We assume basic knowledge of GANs.</p><p>1) Optical Proximity Correction</p><p>With the improvement of semiconductor technology and the scaling down of ICs, traditional OPC techniques are becoming more and more complicated and time-consuming. Yang et al. proposed a new OPC method based on generative adversarial networks (GANs). A Generator (G) is used to generate the mask pattern from the target pattern, and a discriminator (D) is used to calculate the quality of the generated mask. GAN-OPC can avoid complicated computations in ILT-based OPC (inverse lithography technique), but it faces the problem that the algorithm is difficult to converge. To deal with this problem, ILT-guided pre-training is suggested. In the pre-training stage, the Discriminator network is replaced with the ILT convolution model, and only the G network is trained. After pre-training, the ILT model that has very large cost is removed, and the full network is trained. The training flow of GAN-OPC and ILT-guided pre-training is shown in the figure below. The experimental results show that the GAN-based methodology can accelerate ILT based OPC significantly and generate more accurate mask patterns.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/545/1*YkLIbJeuJz2Ku_gamgBxrg.png" /><figcaption>(a) GAN-OPC and (b) ILT guided pre-training</figcaption></figure><p>As an improvement to this, the Enhanced GAN-OPC framework was introduced, which significantly improves the training efficiency in a more elegant way compared to pretraining with ILT engine. The EGAN-OPC framework includes a U-Net structure that allows gradients to be easily backpropagated to early layers and an SPSR architecture for better generated mask quality.</p><ol><li>U-Net: We have noticed the GANs are typically deeper than traditional neural networks, which brings more challenges due to a longer gradient backpropagation path. A common solution is creating shortcut links that allow addition or stacking of feature maps in different layers, such that gradients can be more efficiently backpropagated from output layer to early layers. Here, we enhance our generative design with a U-Net-like structure where intermediate feature maps in the encoder are stacked at corresponding layers in the decoder. Such architecture has two good properties: 1) the inevitable information loss in strided convolution layer can be drastically reduced and 2) the gradient vanishing problem can be alleviated with multiple shortcut links bypassing intermediate feature maps.</li><li>Subpixel Super-Resolution: In the previous designs, low level features in intermediate generator layers are cast back to mask images by standard strided deconvolution operation. SPSR is an upsampling solution that has been widely used in super-resolution tasks.</li></ol><p>Thus, in this article we took a look at a GAN-based mask optimization flow.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c82dcad6881" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/mask-optimization-using-gan-3c82dcad6881">Mask Optimization using GAN</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[High Level Synthesis in VLSI]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/high-level-synthesis-in-vlsi-3deb50603bb7?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/3deb50603bb7</guid>
            <category><![CDATA[high-level-synthesis]]></category>
            <category><![CDATA[vlsi]]></category>
            <dc:creator><![CDATA[Harshada Belgi]]></dc:creator>
            <pubDate>Fri, 19 Mar 2021 05:14:38 GMT</pubDate>
            <atom:updated>2021-03-19T05:14:38.831Z</atom:updated>
            <content:encoded><![CDATA[<p>With the advancements in the digital electronic technologies and devices that are integrating System-on-chip (SoCs), the design complexity of hardware is also increasing.</p><p>The designing of hardware systems is thus being optimised with various methodologies and tools that can produce the design output with reliability and efficiency.</p><p><strong>Levels of abstraction :</strong></p><p>Hardware design can be implemented through various levels of abstraction, or operational levels. Most commonly utilised levels of abstraction are the gate level, register-transistor level (RTL) and algorithmic level.</p><p>In software design, as the program complexity and size grew, it became necessary over a period of time to shift from translators assemblers to compilers, that is, move to higher levels of abstraction to simplify the process of design.</p><p>Similarly, automated hardware synthesis tools were developed due to the translation from higher levels of abstraction for software. Hardware circuits were able to be produced automatically synthesizing an easier description of the required circuit. This automation gave rise to the concept of logic synthesis and high level synthesis.</p><p><strong>About High Level Synthesis :</strong></p><p>While logic synthesis operates with the RTL description of the design, high level synthesis works at a higher level of abstraction, the algorithmic level. High-level Synthesis (HLS) can be defined as the translation from a behavioural description of the intended hardware circuit into a structural description similar to the compilation of higher programming languages like C, C++ into hardware description languages like Verilog or VHDL. HLS acts as the next step in design automation, succeeding logic synthesis.</p><p>The objective of HLS is to extract parallelism from the input description, and construct a micro architecture that is faster and cheaper than simply executing the input description as a program on a processor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/408/1*ncOe6X7P_BxxV4-rRUQHxw.png" /><figcaption>Hardware Development Process</figcaption></figure><p>The input to HLS involves a set of constraints or rules that are set by the designer in order to achieve RTL synthesis.</p><p>The output of this High Level Synthesis can include the following:</p><p><em>RTL Implementation :</em> It includes the RTL netlist which contains all the libraries, parameter specifications, control logic and so on.</p><p><em>Analysis Feedback :</em> It includes the reports on the performance of the synthesis.</p><p><strong>Design steps involved in high level synthesis :</strong></p><p>The following tasks are executed by HLS tools :</p><p>1. Compiling the specification : involves code optimization</p><p>2. Allocating the hardware resources, may it be the storage components or the buses or the ICs : defines the type and number of hardware resources</p><p>3. Scheduling the operations : operations are scheduled in cycles</p><p>4. Binding the scheduled operations to their functional units</p><p>5. Binding the variables to the storage elements</p><p>Generating the RTL architecture : Once allocation, scheduling and binding are done, all the design decisions generate an RTL model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/548/1*ppt7X8MqOYtWllwScMGauw.png" /><figcaption>Design Steps in HLS</figcaption></figure><p>The tasks are managed in a sequential manner which manages the complete computational complexity of the synthesis.</p><p><strong>How is it beneficial?</strong></p><p>As compared to logic synthesis which occurs at the RTL, high level synthesis provides a faster and high-quality path and enhancing the design productivity.</p><p>The benefits that HLS provides are derived from the fact that it operates at a higher level of abstraction with high level languages :</p><p>1. It easily evaluates algorithmic changes.</p><p>2. It can explore multiple architectures while focusing on the designing of core functionalities.</p><p>3. It can easily debug and test functionality of the input descriptions.</p><p>The utilization of HLS however depends majorly on whether the micro architecture given can be satisfied with designing in RTL or if different algorithms and architectures can be explored to experiment under a variety of constraints. There have been multiple areas that have benefitted from HLS like automation industries, image processing, FPGA based applications and so on.</p><p><strong>References :</strong></p><ol><li>An Introduction to High-Level Synthesis, Philippe Coussy, Michael Meredith, Daniel D. Gajski, Andres Takach</li><li>High-level Synthesis, Issam W. Damaj, Dhofar University</li><li>Understanding high-level synthesis design’s advantages, By John Sanguinetti Chief Technology Officer Forte Design Systems Inc.</li><li>Machine Learning for Electronic Design Automation: A Survey</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3deb50603bb7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/high-level-synthesis-in-vlsi-3deb50603bb7">High Level Synthesis in VLSI</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Logic Synthesis using Deep Neural Networks]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/logic-synthesis-using-deep-neural-networks-5c289bb962b2?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/5c289bb962b2</guid>
            <category><![CDATA[deep-neural-networks]]></category>
            <category><![CDATA[vlsi]]></category>
            <dc:creator><![CDATA[Anmol Salvi]]></dc:creator>
            <pubDate>Wed, 03 Mar 2021 04:06:54 GMT</pubDate>
            <atom:updated>2021-03-03T04:06:54.477Z</atom:updated>
            <content:encoded><![CDATA[<p>Logic Synthesis using Deep Neural Networks</p><p>INTRODUCTION: <br>With advancements and growth in the data science industry , the current applications of DNN(Deep Neural Networks) and various other machine learning techniques are seemingly limitless . Similarly in this manner , DNN also finds it’s application in logic synthesis and can be used to better optimize the entire process and also reduce the time consumed in this process. Before looking into how DNN techniques are applied for logic synthesis applications, first certain key terms need to be discussed. They include:-<br>A)Design Space Complexity of ALS(Approximate Logic Synthesis):- This involves gate replacement by going through each node and performing approximation on all and also calculating the error rate at primary outputs resulting from this approximation. Thus it’s complexity analysis has 2 parts namely:-<br>1)Node replacement <br>2)Error propogation<br>B)Deep Neural Networks:-A DNN has an input layer along with a number of hidden layers which are basically a group of neurons and are used for feature selection. An output layer also exists. Inputs from the hidden layer are passed through a non linear activation function which includes the relu , leaky relu and sigmoid functions among others and they help minimize error in prediction using backpropogation and updation of features at important nodes.<br>IMPLEMENTATION:-The input circuit to be optimized is first mapped using the cut-based technology mapping module of ABC post which depending on the mode of optimization (eg-power), critical nodes are extracted and are then passed onto the approximation module , this module replaces a node and after consulting the trained DNN model predicts the worst error rate at a primary output. If this worst error rate doesn’t violate the already predetermined error constraint by the user , the approximation module accepts this replacement , else it undo’s the last replacement and moves onto another node. This approximation process ends when either the error budget is consumed fully or all nodes in the circuit are visited. The learning rate needs to be set to a small value and using an optimizer such as Adam is preferred. The number of epochs can be determined by checking behaviour and accuracy of the validation set , a number greater than 10 is preferred to start at. Binary cross entropy is preferred to be used as the loss function.<br>ADVANTAGES:-A)Saves time.<br> B)Ensures greater accuracy.<br> C)Thereby ensures better performance.<br> <br>CONCLUSION:-One can thus conclude that Logic synthesis using DNN is one of the many boons of the advancement of the data science industry and making the most of this technique in order to better optimize power consumption &amp; time consumption parameters among others , is essential and can thus serve to revolutionize just how logic synthesis is carried out in the VLSI industry.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5c289bb962b2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/logic-synthesis-using-deep-neural-networks-5c289bb962b2">Logic Synthesis using Deep Neural Networks</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Reinforcement learning for logic synthesis]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/reinforcement-learning-for-logic-synthesis-9e3a9dab3b14?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/9e3a9dab3b14</guid>
            <category><![CDATA[reinforcement-learning]]></category>
            <dc:creator><![CDATA[ISHAN APHALE]]></dc:creator>
            <pubDate>Sun, 28 Feb 2021 16:57:01 GMT</pubDate>
            <atom:updated>2021-02-28T16:57:01.023Z</atom:updated>
            <content:encoded><![CDATA[<p>Logic synthesis is one of the most important steps in modern chip design, and consequently in EDA. Logic synthesis converts the behavioural level description into gate level description, which is one of the most important problems in EDA. Logic synthesis is the implementation of the specific logic functions by generating a combination of gates selected in a given cell library, and optimizes the design for different goals. Being a complicated process, it cannot be solved perfectly and so heuristic algorithms are widely used in this stage, which include lots of ML methods.</p><p>The emergence of new technologies and slowing down of Moore’s law is putting increasing pressure on the field of EDA. Logic synthesis requires extensive tuning of the synthesis optimization flow where the quality of results (QoR) is dependent on the sequence of optimizations used. Efficient design space exploration is challenging because of the exponential number of possible optimization permutations. There is a constant need to improve optimization algorithms, and it is necessary for these methods to be autonomous.</p><p>Here we examine two approaches for the same. Both approaches are given a Directed Acyclic Graph (DAG) as an input.<br> 1) DRiLLS (Deep Reinforcement Learning based Logic Synthesis):</p><p>DRiLLS, which stands for Deep Reinforcement Learning-based Logic Synthesis, effectively maps the design space exploration problem to a game environment. Unlike most reinforcement learning environments where gamification drives the behavior of the environment, the task here involves combinatorial optimization the design of a given circuit. This makes it very challenging to not just define the state of the game (i.e. environment), but the long term incentives for an agent to explore the design space and not to fall into local minimums.</p><p>This method maps the complex search space to a “game” where an advantage actor critic (A2C) agent learns to maximize its reward (reduce area subject to a delay constraint) by iteratively choosing primitive transformations with the highest expected reward. There is a multi-objective reward function defined that takes into account the change in both the design area and the delay. In particular, the agent is rewarded for reducing the design area, while keeping the delay under a pre-defined constraint value.</p><p>2) GCN (Graph Convolution Networks):</p><p>This method suggests that the process of logic network optimization can be cast as a deterministic Markov Decision Process (MDP). Such a process can also be thought of as a single-player game of perfect information. The game consists of states and moves that transition from one state to the next one. The different game states correspond to logic networks. Moves in the game correspond to operations on the networks. In this approach, they use the majority inverter graphs (MIGs) as logic network data structures. MIGs correspond closely to a median (ternary majority) algebra, which allows to define a sound and complete move set. Such a move set allows it to reach any point in the design space. The policy function has been modelled as a deep neural network, based on Graph Convolution Networks (GCN). For every node in the MIG node features should be derived, based on a local neighbourhood, that capture information that is useful towards predicting which move is beneficial to apply.</p><p>Notice how both approaches take the form of a game. This is the beauty of reinforcement learning, where an agent is not told what action it should take, but instead receives a reward or penalty for actions. Autonomous learning of optimization is important and RL is an important step ahead.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9e3a9dab3b14" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/reinforcement-learning-for-logic-synthesis-9e3a9dab3b14">Reinforcement learning for logic synthesis</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Logic Synthesis in VLSI]]></title>
            <link>https://medium.com/logic-synthesis-in-vlsi/logic-synthesis-in-vlsi-2132ecf5e1e9?source=rss----84dd639248b1---4</link>
            <guid isPermaLink="false">https://medium.com/p/2132ecf5e1e9</guid>
            <dc:creator><![CDATA[PRANAV BAPAT]]></dc:creator>
            <pubDate>Fri, 26 Feb 2021 04:39:48 GMT</pubDate>
            <atom:updated>2021-02-26T04:39:48.073Z</atom:updated>
            <content:encoded><![CDATA[<h4>What is Logic Synthesis in VLSI?</h4><p>Logic synthesis is basically a process through which a specification of a desired circuit behavior is turned into or translated into a design implementation in terms of logic gates. This translation of a design representation into a circuit comprising of transistors is known as logic synthesis. It is a major part in any circuit designing for implementation of hardware. This synthesis is generally done by using a computer program called as a synthesis tool. The designs specified in hardware description languages are translated using these synthesis tools. The specification of circuit behavior is typically given at register transfer level. Thus, logic synthesis is basically converting a high-level description of design into an optimized gate-level representation.</p><p>The logic synthesis can be majorly summarized in three main points that are translation, optimization and technology mapping.</p><p>Translation: In this, the RTL code is translated to a representation which is technology independent. The logic that is converted is then available in a boolean equation form.</p><p>Optimization: The obtained Boolean equation is optimized using sum of products (SOP) optimization methods.</p><p>Technology mapping: The optimized Boolean equations are then mapped to technology dependent library logic gates based on design constraints and library of available technology gates.</p><h4>Why perform Logic Synthesis?</h4><p>There are various advantages of performing logic synthesis.</p><p>1. Automatically manages many details of the design process due to which there are fewer bugs and the productivity is also improved.<br>2. It abstracts the design data that is the description given in HDL from any particular implementation technology. The designs can be re-synthesized targeting various chip technologies.<br> E.g: first implement in FPGA then later in ASIC<br> 3. In some cases, leads to a more optimal design than could be achieved by manual means (logic optimization)</p><h4>Logic Synthesis Flow</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/265/1*5CmQy65lnykXfD7CoBVaRA.jpeg" /></figure><p><strong>RTL description:</strong><br> It basically describes the design at a high level using Register Transfer Level constructs.<br> <br> <strong>Translation:</strong><br> The synthesis tool converts the RTL description to an un-optimized Internal representation, that is we get the Boolean form.<br> <br> <strong>Un-optimized Intermediate Representation:</strong><br> It is represented internally by the logic synthesis tool in terms of internal data structures.</p><p><strong>Logic Optimization:</strong><br>Logic is optimized to remove redundant logic and use fewer hardware.</p><p><strong>Technology Mapping and Optimization:</strong><br> The synthesis tool takes the internal representation and implements the representation in gates, using the cells provided in the technology library.</p><p><strong>Technology Library:</strong><br> It consists of library cells that can be basic gates or macro cells.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2132ecf5e1e9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/logic-synthesis-in-vlsi/logic-synthesis-in-vlsi-2132ecf5e1e9">Logic Synthesis in VLSI</a> was originally published in <a href="https://medium.com/logic-synthesis-in-vlsi">logic-synthesis-in-vlsi</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>