White Paper Big Data Analytics Pilot, Internet of Things Manufacturing

Optimizing Manufacturing with the Internet of Things

Intel manufacturing advances operational efficiencies and boosts the bottom line with an IoT and big data analytics pilot

Introduction proves IoT success The volume, variety, and velocity of data1 being produced in manufacturing is growing exponentially, creating opportunities for those diligently analyzing this by using big data analytics data to gain a competitive edge, respond to changing market dynamics, and to bring cost savings, increase manufacturing margins, productivity, and efficiency. Equipment across the factory floor is generating thousands of different data types, such as unit predictive maintenance, production data at multiple levels, equipment operation data, process data, and and higher product yields human operator data, which can be stored to be made available for analysis.

to its own manufacturing Although large manufacturers designs for manufacturability, thereby processes. have been using statistical process improving supply chain management control and statistical data analysis and introducing the use of customized to optimize production for years, the manufacturing services to shorten time extreme composition of today’s data to market for products customized provide opportunities to deploy new for smart consumers across various approaches, infrastructure, and tools. geographies. Manufacturing industries are ready to This paper outlines an Internet of embrace the use of big data, supported Things (IoT) pilot in one of Intel’s by higher compute performance, open manufacturing facilities to show how standards, the availability of industry data analytics applied to factory know-how, and the vast availability equipment and sensors can bring of highly skilled, data statisticians operational efficiencies and cost fueled by academics. With access savings to manufacturing processes. to new manufacturing intelligence, With industry collaboration from manufacturers will improve quality, Cloudera, Dell, Mitsubishi Electric, increase manufacturing throughput, and , this IoT big have better insights into root cause data analytics project is forecasted to of manufacturing issues, and reduce save millions of dollars annually along machine failure and downtime. with additional return on investment With these new business values and business value. technology capabilities, manufacturers will be able to change business models and practices in order to optimize Optimizing Manufacturing with the Internet of Things

Challenge: How to extract data from manufacturing execution and enforce hierarchies of records and value from all of the data in systems and enterprise systems. On fields within the data. In manufacturing, manufacturing? the other hand, unstructured data the power of big data technology stems such as images, text, machine log from the ability to merge and correlate Big data is characterized by huge files, human-operator-generated shift these data set types to create business data sets with varied data types, reports, and manufacturing social value through newfound insights. The which can be classified as structured, collaboration platform texts may be other value proposition of big data semi-structured, or unstructured, as in a raw format that requires decoding technology is it allows manufacturers to shown in Table 1. Structured data fits before data values can be extracted. aggregate and centralize various types nicely into neatly formatted tables, Semi-structured data is a form of of data in a cost-effective, scalable making it relatively easy to manage structured data that does not conform manner. and process. Structured data has the to the formal structure of data models In manufacturing, process variability advantage of being easily entered, associated with relational databases stemming from various factors, like stored, queried, and analyzed. or other forms of data tables, but material, process recipes and methods, Examples include manufacturing data nonetheless contains tags or other and equipment differences, drives a stored in relational databases, and markers to separate semantic elements real business need for manufacturers to turn to a big data solution based on MANUFACTURING REAL-TIME, a scalable platform that can grow with INDUSTRIES SEMI-STRUCTURED UNSTRUCTURED STRUCTURED their businesses and manufacturing EXAMPLES DATA DATA DATA requirements. Machine data is strongly Semiconductor Machine builder Operator shift RDBMS databases correlated to yield, quality, and output, standards like reports Electronics NoSQL* thereby providing valuable information SECS/ GEM, EDA or Machine logs Solar Enterprise data to proactively detect processes that custom-based on Error logs warehouse Machinery are getting out of control. However, COM, XML Texts Files stored in some types of manufacturing generate Energy Sensors (vibration, manufacturing PCs Vision images massive data files (gigabytes in a few Automobiles pressure, valve, and Spreadsheets days per tool type, as shown in Table 2), acoustics), Relays Audio/Video Aerospace limiting the ability to store, analyze, and RFID Manufacturing Chemical and collaboration social extract useful information from them Pharma Direct from PLCs, platforms using conventional methods. Without Motor and drives, Metalworking the use of big data technologies, it is Direct from motion Food and Beverage extremely hard to even visualize the controllers, robotic Pulp and Paper arm information in large data sets from various sources. Clothings and Manufacturing Textiles historians (time Furniture series data structures) Table 1. Manufacturing data examples

DATA SIZE DATA TYPES (per week) EXAMPLES Machine parameters ~5 GB per machine Used to monitor machine performance: and error logs dispense height, placement (x, y, z), belt speed, flow rate, oven temperature, laser power, etc. Machine events ~10 GB per machine Used to measure process time: start dispense, end dispense, start setup, and end setup Defect images from ~50 MB per unit or Used to identify root cause of failure modes, vision equipment 750 GB per lot defect commonality, defect mapping Table 2. Data size examples

2 Optimizing Manufacturing with the Internet of Things

Addressing the challenge: IoT pilot using big data analytics server and IoT gateway in Intel manufacturing

BIG DATA ANALYTICS BUSINESS INTELLIGENCE AND VISUALIZATION

Data Analytics (Visualization, Monitoring, Statistical) CORPORATE INTRANETS

VPN

FIREWALL Automation DBs DATA INGESTION Data Distill (ETL) DATA ACQUISITION MANUFACTURING SUBNETS

Gateway

Gateway Tools

Data Storage, Retention, and Management

Private Cloud Infrastructure Testers

Figure 1. Building blocks for end-to-end infrastructure enabling manufacturing intelligence from the factory floor to the datacenter

Figure 1 shows a high-level IoT As an example, the architecture be accessed with other analytical or manufacturing architecture for small can: clean, extract, transform, and monitoring applications on the network. to large data sets that constitute data consolidate structured data from Other enhanced capabilities could acquisition and aggregation of various existing databases and unstructured include running analytics in Hadoop types of data from the manufacturing data from tool sensors and log files or other types of file systems, or shop floor and the manufacturing from legacy equipment in a data store running analytics in memory for faster network, which opens up the platform (i.e., Hadoop*). The data can performance. The results of the analysis possibilities of visualizing, monitoring, then be visualized and analyzed by can be presented to users via intuitive and for new business high-level factory applications running visualization capabilities in the business intelligence. in different virtual machines on the intelligence layer of the network. same on-premise server as the data store server. Alternatively, the data can

3 Optimizing Manufacturing with the Internet of Things

Big data analytics IoT server setup in Intel manufacturing pilot

Revolution Enterprise* from Revolution Analytics Business Business

Intelligence By linking with high-performance Intel® MKL multithreaded math libraries, Revolution R

Revolution Analytics DeployR* Web Services Enterprise uses the power of

Data Existing Factory multiple processors to accelerate Analytics Applications Consolidated into common matrix calculations. In Virtual Machines R Based Manufacturing Analytics Template 1, 2, and 3 addition, Revolution R Enterprise ETL parallel, distributed algorithms enable statistical analysis of Big Cloudera AquaFold* Data to scale linearly with multi-

Batch Interactive Machine core servers, high-performance Search Processing SQL Learning computing clusters, big data MonetDB* appliances, and Hadoop*.

Management Resource Management

Data Store and Metadata Storage Integration PostgreSQL* • Revolution R Enterprise* from Revolution Analytics is analytics Enterprise Linux* for Virtual Datacenters software built upon the powerful Dell PowerEdge* VRTX open source R statistics language. The software provides a seamless, secure data bridge between analytics OS VirtualizationOS solutions and enterprise software, thereby solving a key integration problem faced by businesses Figure 2. Big data analytics server adopting R-based analytics alongside existing IT infrastructure.3 The big data analytics server used last-level (L3) cache, and up to 1.5TB by Intel manufacturing in its pilot maximum memory capacity, along with • MonetDB*: an open source column- deployment is shown in Figure 2. communication pathways to move data oriented database management A compact system from Dell, the more quickly. system designed to provide high PowerEdge* VRTX,2 was selected to performance on complex queries The two M820 servers host analytics host the big data and analytics software against large databases, such as and application software, and the in a private cloud setting as an on- combining tables with hundreds Hadoop nodes, which run in multiple premise server. The system consists of of columns and multimillion rows. virtual machines. Red Hat Enterprise a Dell PowerEdge VRTX chassis with MonetDB has been applied in high- Linux* for Virtual Datacenters twenty-five 900GB hard disks and two performance applications for data operating system provides a complete Dell PowerEdge M820 blade servers, mining, online analytical processing virtualization software solution for each equipped with four Intel® Xeon® (OLAP), geographic information servers designed for a scalable and fully processors from the E5-4600 product systems, and streaming data virtualized datacenter. 4 family on each blade. The Intel Xeon processing. processor E5-4600 product family Analytics and Application node • PostgreSQL*: a powerful, open provides a dense, cost-optimized, Figure 3 shows how the software is source object-relational database four-socket processor solution with allocated to different virtual machines system used for online transaction 5 up to eight cores, up to 20MB of (VMs). The node hosts five VMs that run processing (OLTP). the various analytics and application workloads. This includes: 4 Optimizing Manufacturing with the Internet of Things

• AquaFold*: an application server used to build and deploy production Column-store quality database and reporting database for fast applications quickly. It includes Revolution Analytics DeployR* Web Services analytics workloads capabilities like multiple sources data RDBMS/row-store Manufacturing Manufacturing Manufacturing synchronization, cross-database data Analytics Analytics Analytics database for OLTP Template 1 Template 2 Template 3 workloads migration, data warehousing extract transform and load, scheduling data exports/ imports, and custom Analytics and Application node business intelligence dashboards. Dell M820 (four 8-core processors, 256GB RAM) Supports communication to various production databases in the network CLOUDERA ENTERPRISE* DATA HUB like PostgreSQL, SQL*,

MANAGEMENT Oracle*, and IBM DB2* with ODBC* 6

BATCH ANALYTIC SEARCH MACHINE STREAM 3RD PARTY DATA and JDBC* connections. PROCESSING SQL ENGINE LEARNING PROCESSING APPS Hadoop* node WORKLOAD MANAGEMENT Four virtual machines are provisioned to run a , four-node Cloudera Hadoop cluster, consisting of one head STORAGE FOR ANY TYPE OF DATA MANAGEMENT

UNIFIED, ELASTIC, RESILIENT, SECURE SYSTEM node and three worker nodes. • Apache Hadoop is an open source FileSystem Online NoSQL distributed software platform for scalable distributed computing. Written in Java, it runs on a cluster of industry-standard servers configured Hadoop* Node with direct-attached storage and Dell M820 (four 8-core processors, 512GB RAM) cost-effectively scales performance by adding economical nodes to the HDD: 900GB 10K RPM SAS 6GBps 2.5 cluster. • Cloudera Enterprise Data Hub* (CDH) offers a unified platform for big data by providing one place to store, process, and analyze all of their data resulting in an extension of existing investment value and enabling fundamental new ways to derive value from their data. CDH is 100 percent Apache licensed open source and is unique in offering unified batch processing, interactive 10 x 900GB ~ 6TB 15 x 900GB ~ 10TB usable SQL, and interactive search and role- 7 Figure 3. Software allocations to virtual machines on the big data analytics server based access controls.

5 Optimizing Manufacturing with the Internet of Things

collected from sensors or via the of tests required could be reduced, network when supporting sophisticated resulting in decreased test time. system control and operations. At the Implemented as a proof of concept, this A closer look at the core of this controller is a hardware solution avoided test costs of $3 million Cloudera stack platform based on Intel® architecture in 2012 for one series of Intel® Core™ Cloudera Enterprise Data and the Wind River VxWorks* real-time processors. Extending this solution Hub* (CDH) delivers the core operating system. to more products, Intel expects to realize an additional $30 million in elements of Hadoop*—scalable Mitsubishi Electric developed C language cost avoidance.9 storage and distributed controllers of MELSEC-Q Series to satisfy computing—along with the diverse requirements of factory Improved manufacturing monitoring additional components such as automation, including excellent reliability, a user interface, plus necessary Data-intensive processes also help tolerance of harsh environments, and Intel detect failures in its manufacturing enterprise capabilities such as long-term availability. These features security, and integration with a line, which is a highly automated make it a robust and reliable product environment. Intel pulls log files out broad range of hardware and that requires little maintenance for IoT software solutions. of manufacturing tools and testers manufacturing applications. across the entire factory network, CDH can help businesses In place of ladder logic used in which can be up to 5 Terabytes an transform and accelerate the conventional programmable logic hour. By capturing and analyzing this way big data is used. When controllers, the C language controllers information, Intel can determine combined with datacenter of MELSEC-Q series use the when a specific step in one of its architecture based on Intel® international standard C languages manufacturing processes starts to Xeon® processors, Cloudera and (C and C++) for greater programming deviate from normal tolerances.10 Intel provide a comprehensive flexibility. This allows users to take full approach for industrial and Specific to the current end-to- advantage of their existing C language end platform pilot deployment manufacturing businesses software and development. seeking to analyze data to make discussed here, Intel, in collaboration factories run more efficiently. CIMSNIPER* is a data acquisition with Mitsubishi Electric, Cloudera, and processing software package Revolution Analytics, and Dell, for Mitsubishi Electric C Language successfully pioneered capabilities Controller of MELSEC-Q series. It that have made tremendous headway Internet of Things (IoT) Gateway can collect process data (including in the use of data mining sciences An Intel® Atom™ processor-based SECS messages) and manufacturing to solve practical manufacturing gateway by Mitsubishi Electric, called equipment errors without modifications issues, thus saving Intel millions of the Mitsubishi Electric C Language of existing systems. dollars through cost avoidance and Controller of MELSEC-Q series*,8 improved decision making. The major is used to aggregate and securely Big data analytics cases at goal of this project is to extract the ingest data into the big data analytics Intel manufacturing value propositions of data and data server. Data ingestion is the process of analytics to obtain better insights validating, filtering, and reformatting Over the past two years, Intel has in predictive manufacturing and the data to make it easier for the big developed more than a dozen big reduce manufacturing costs without data analytics software to work on it. data projects that have bolstered both sacrificing throughput or quality. operational efficiency and the bottom The Mitsubishi Electric C language line. Here are a couple of examples: The following details some of controllers of MELSEC-Q series are the groundbreaking work and embedded solutions equipped with Reduced product test times discoveries Intel accomplished numerous features characteristic of Every Intel® chip produced undergoes through the integration of big intelligent systems, including robust a thorough quality check involving a data analytics and technologies network connectivity and the high complex series of tests. Intel found that for the IoT in manufacturing. computational performance needed by using historical information gathered to process large amounts of data during manufacturing, the number

6 Optimizing Manufacturing with the Internet of Things

Pilot Results: yield losses by up to 25 percent.12 In Use Case 3: Using image analytics to Use Case 1: Reducing non-genuine addition, Intel was able to save spare identify good or defect units production yield loss through the cost by reducing the need to replace Background monitoring and analysis of machine spare parts before they fail during A machine vision equipment is a module parametric values and timely preventive maintenance resulting in an that screens units and categorizes them replacement of parts before they fail estimated 20 percent reduction in spare into good and marginal units. The good cost. Background units are sent forward to be processed Automated Test Equipment (ATE) is a Use Case 2: Reducing yield losses by while the marginal ones are inspected machine that is designed to perform eliminating and minimizing incorrect and determined by a manufacturing tests on different devices referred to ball assembly in ball attach equipment specialist to be either good or defective. as a devices under test (DUT). An ATE This manual process takes time. Background uses control systems and automated The ball attach module is where solder information technology to rapidly Problem Statement paste is printed onto the lands on the perform tests that measure and The manual process to inspect and substrates. Solder balls are placed evaluate a DUT.11 The ATE systems categorize marginal units is cumbersome into the ball attach lands, and the interface with an automated placement and sometimes takes about 8 hours to paste holds them in place. The entire tool, called a handler, that physically successfully segregate a host of true package goes through a reflow oven, places the DUT on a Test Interface Unit reject units from marginal ones. which melts the paste and balls to the (TIU) so that it can be measured by substrate lands. This is caused by the time it takes the equipment. for the units to reach the operator, Solder balls are vacuumed onto tiny flow to a segregation module, and Problem Statement holes of the placement head. The head to ultimately be segregated. Image Defective TIUs will wrongly categorize is inspected for excess or missing analytics enables identification of good units as bad, which negatively balls. Once the head is aligned to the reject units moments after being impacts Intel manufacturing operation substrates, the balls are placed in the inspected by the inspection module. costs. Defective TIUs can cause DUTs solder paste on the substrates. After to be wrongly categorized, including releasing the balls, the placement head Results and Benefits the rejection of good units. Intel is inspected for any remaining balls. The marginal images recorded at the manufacturing’s objective was to Finally, the substrates are inspected by machine vision equipment module detect TIU defects before they happen a camera vision system for any missing are preprocessed. Each image, which so they could repair or replace them or shifted balls. is unstructured data, is resized, before units are wrongly categorized. cropped, converted into grayscale When a faulty TIU wrongly categorizes Problem Statement before converting each pixel to good units as bad, the units are Units with missing balls are faulty binary. The next stage of the process scrapped. Some components were material and a yield loss. There are involves feature selection, where the replaced with spare parts during numerous scenarios in which balls unstructured image is defined by a set regular preventive maintenance, even are missing from the units, including of distinct values. These values are if they were still operating properly inadequate vacuum pressure. then fed into various machine learning to avoid such problems. algorithms to determine true rejects Results and Benefits and marginal rejects. Results and Benefits Yield By visualizing and correlating sensor The analytics capability predicted up readings with various machine data The image analytics shortens the to 90 percent of potential TIU failures and execution system data, Intel was time to save true rejects from a pool before being triggered by the existing able to reduce yield losses, optimize of marginal units. Image analytics factory’s online process control system. maintenance cost, and avoid sudden identifies defects roughly 10 times This helped, in this case, to replace equipment downtime.12 This enabled faster than the manual method.12 defective TIUs before causing over the technicians to proactively fix the rejection of good units, thus reducing problem, toward the journey of having a predictive maintenance capability.

7 Optimizing Manufacturing with the Internet of Things

DATA SOURCES DATA VISUALIZATION ON EQUIPMENT

Sensors Business Intelligence, Visualization, Tools and Apps

VALUE DATA ANALYTICS Predictive Maintenance, Reduced MTTR, Revo DeployR WebServices Improve Cycle Time, Reduce Cost R Based Manufacturing Analytics Template 1, 2 and 3 DATA FILTERING AND DATA TRANSPORT

DATA PREPARATION FTP/ MQTT/ RESTFul Extract Transform Load, Database Extraction Tools, SQL, Phyton, Perl

Scoop DATA STORAGE/WAREHOUSE

Operational Data Cloudera CLOUDERA ENTERPRISE* DATA HUB

Hadoop* MANAGEMENT

BATCH ANALYTIC SEARCH MACHINE STREAM 3RD PARTY DATA (MapReduce PROCESSING SQL ENGINE LEARNING PROCESSING APPS and HDFS) WORKLOAD MANAGEMENT RDMS/ OLTP DATA DISTILLATION STORAGE FOR ANY TYPE OF DATA MANAGEMENT

UNIFIED, ELASTIC, RESILIENT, SECURE SYSTEM

FileSystem Online NoSQL NoSQL COL/ Phyton, R, Perl, Excel (Hadoop) OLAP Various Lib

Figure 4. High-level data flow

Data Flow • All incoming factory data is stored - Flume* can receive a stream of in Hadoop*. The following, non- continuous log data. Figure 4 shows how data flows from the exhaustive list shows some of the previously discussed use cases. • The data specific to these use cases possibilities that can be achieved described in the section above are in • Machine data and sensors data are using readily available capabilities in the form of Comma-Separated Value sent by the Intel Atom processor- Hadoop: (CSV) files or raw images. While the based gateway to the Big Data - HTTP: The Big Data Analytics Hadoop ecosystem includes plenty Analytics Server (BDAS) as they are Server exposes an authenticated of ways to ingest data (as described acquired in real time. For instance, REST HTTP endpoint that supports in the previous point), some of the machine data from the ball attach operations into HDFS. machines in the factory have limited module acquired over machine network transfer capabilities, which interface ports and from analog - Apache Sqoop* provides the required custom engineering to sensors are streamed at a velocity of connectivity tool for moving data deliver data into HDFS: a few MBs per a few seconds. from non-Hadoop data stores— such as relational databases and data warehouses—into Hadoop.

8 Optimizing Manufacturing with the Internet of Things

- FTP: The IOT gateway has an FTP Summary of key learnings client that periodically connects to and conclusions the Big Data Analytics Server and Intel integrated and validated a big Big data analytics and transfers over the latest acquired data analytics on-premise server the Internet of Things in data directly into HDFS. Other solution with data extracted from manufacturing as an end-to- streaming protocols like MQTT Intel’s own manufacturing network end platform is the critical and REST can be used depending and from equipment and sensors backbone to enable the vision on the real time streaming and through the use of Internet of Things of smart manufacturing. analytic requirements. gateway to validate the business value This platform is scalable - CIFS share (Windows share): of Internet of Things in manufacturing. and available in various The Big Data Analytics Server The Mitsubishi Electric C language configurations using currently provides a Windows/CIFS share controllers of MELSEC-Q series available industry-standard directory that the gateway can were used. The pilot involved close building blocks. copy files into. collaborations between the factory engineers, the IT department, and • The CSV files are directly imported industry experts from Cloudera, Dell, into HBase* using Pig, while the raw For Intel, this pilot is forecasted to Mitsubishi Electric, and Revolution images first go though a preprocess save millions of dollars (USD) annually Analytics. The team started leveraging using a map-reduce job using with additional return on investment existing machine performance and computer vision techniques to business values that Intel is still monitoring data, then proceeded produce a text data representation realizing. Benefits include improving to leverage big data analytics and of the image. equipment component uptime, modelling to ingest the data to predict minimizing wrong classification of • Depending on the operational potential excursions and failures. Being good units as bad (hence increasing requirements, the data is stored able to predict the machine component yield and productivity), enabling in one of three databases: NoSQL failures allows engineers to repair and predictive maintenance, and reducing (Hadoop), RDMS/SQL, or Coli/OLAP. prevent the excursion, hence reaping component failures. There are many tremendous savings from wasting • The databases are accessed and other types of parametric, metrology, production units, time to repair, and processed using various tools, product, and equipment data—both machine components. including AquaFold, ad hoc reporting, structured and unstructured within workflow scheduling, and ETL and An integrated suitcase of various Intel manufacturing environment and database integration. At the same software building blocks on the Big Data machines—which could be mined time, the Cloudera Distribution for Analytics Server and IOT gateway were and analyzed to extract new business Hadoop performs various operations used. This framework can be applied values. Capitalizing on this opportunity on the data. and implemented for manufacturers will enable further efficiency and which have not started leveraging on the productivity enhancements to the • The processed data is further intelligence contained in manufacturing factory and ultimately create a analyzed using Revolution data. Manufacturers that have already competitive advantage. Analytics tools designed for been using data to improve their factory applications. efficiency can incrementally add to their • The data is presented to users in easy existing capabilities to evolve their data to understand dashboards. mining and analytics capability to the next level.

9 1. Gartner analyst Doug Laney introduced the 3Vs concept in a 2001 MetaGroup research publication, 3D data management: Controlling data volume, variety and velocity. http://blogs.gartner.com/doug-laney/files/2012/01/ad949-3D-Data-Management-Controlling-Data-Volume-Velocity-and-Variety.pdf. 2. www.dell.com/us/business/p/poweredge-vrtx/pd 3. www.revolutionanalytics.com/revolution-r-enterprise 4. en.wikipedia.org/wiki/MonetDB 5. www.postgresql.org/about 6. www.aquafold.com/ 7. www.cloudera.com/content/cloudera/en/products-and-services/cloudera-enterprise.html 8. www.mitsubishielectric.com/fa/products/cnt/plcq/items/index.html 9. Source: “2012-13 Intel IT Annual Report,” www.intel.com/content/www/us/en/it-management/intel-it-best-practices/intel-it-annual-performance-report-2012-13.html. 10. Source: “Intel Cuts Manufacturing Costs With Big Data,” InformationWeek, March 18, 2013, www.informationweek.com/software/information-management/intel-cuts- manufacturing-costs-with-big-data/d/d-id/1109111. The TCO or other cost reduction scenarios described in this document are intended to enable you to get a better understanding of how the purchase of a given Intel product, combined with a number of situation-specific variables, might affect your future cost and savings. Nothing in this document should be interpreted as either a promise of or contract for a given level of costs. 11. Source: http://www.techopedia.com/definition/2148/automatic-test-equipment-ate 12. Results might vary depending on package size, process, and equipment used in the manufacturing process. INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL® PRODUCTS. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. EXCEPT AS PROVIDED IN INTEL’S TERMS AND CONDITIONS OF SALE FOR SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER, AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO SALE AND/OR USE OF INTEL PRODUCTS INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. UNLESS OTHERWISE AGREED IN WRITING BY INTEL, THE INTEL PRODUCTS ARE NOT DESIGNED NOR INTENDED FOR ANY APPLICATION IN WHICH THE FAILURE OF THE INTEL PRODUCT COULD CREATE A SITUATION WHERE PERSONAL INJURY OR DEATH MAY OCCUR. Intel may make changes to specifications and product descriptions at any time, without notice. Designers must not rely on the absence or characteristics of any features or instructions marked “reserved” or “undefined.” Intel reserves these for future definition and shall have no responsibility whatsoever for conflicts or incompatibilities arising from future changes to them. The information here is subject to change without notice. Do not finalize a design with this information. The products described in this document may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request. Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing your product order. Copies of documents which have an order number and are referenced in this document, or other Intel literature, may be obtained by calling 1-800-548-4725, or by visiting Intel’s Web site at www.intel.com. Copyright © 2014 Intel Corporation. All rights reserved. Intel, the Intel logo, Intel Atom and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries. *Other names and brands may be claimed as the property of others. 0914/KW/CMD/PDF Please Recycle 331218-001US