The Kynix Blog

A bump circuit with flexible tuning ability that uses 500 times less power and is smaller than previous circuits has been demonstrated by researchers at the University of Tennessee in the US."The challenges and requirements of the analogue deep-learning system inspired us to come up with this radically new design," said Junjie Lu, the lead author. "We implemented the bump circuit by preceding the current correlator with a novel nano-power tunable transconductor to achieve variable width and height. By significantly reducing the power consumption of the bump circuit, this work makes possible the realisation of analogue learning and signal processing systems that achieve better energy efficiency than their digital equivalents, and ultimately fully autonomous systems, which are able to get both information and energy from the environment without external intervention."Towards flexible transferThe bump circuit is a family of circuits with bell-shaped, non-linear transfer functions. First appearing in 1991, they are widely used to provide similarity or distance measures in analogue signal processing systems such as support vector machines, neural networks and analogue machine-learning systems.The original bump circuit design lacked the ability to change the width of its transfer function, which is desirable in many applications to represent distributions with different variance or templates with different model parameters. A common approach to solve this is to pre-scale the input voltage, but the circuits required are physically large and consume a lot of power. Other approaches also have limitations such as complex circuitry, large physical size, and a restricted number of possible widths achievable.Hidden depthsThe researchers from the University of Tennessee designed their circuit as an important building block in an analogue deep-learning machine, which is able to perform unsupervised learning and extract salient features from high-dimensional input data, with a much better power efficiency than the existing digital machine learning implementations.Large-scale systems require the computational element, or bump circuit in this case, to be very efficient in both power and area. It is also important that the output features, which are the confidence scores that the current input belongs to each of the previous observations, take both the mean distances and probabilistic variances into account. A bump circuit that has a tunable centre for mean tuning, width for variance tuning, and height for normalisation is therefore highly desirable, and if these three bump parameters can be individually tuned and controlled by a single signal, this would greatly help with on-chip trainability.To achieve the variable height and width, the researchers designed and incorporated a novel transconductor, linearised using the drain resistances of saturated transistors. They adopted a pseudo-differential structure to allow operation with a low supply voltage, and designed a common mode feedback circuit to provide common mode rejection for the pseudo-differential structure to get a tunable bump height.The whole circuit uses 18.9 nW power from 3 V supply which is 1/500 th of the power of the next best bump circuit with tunable width. Implemented in 0.13 µm CMOS, it is smaller in area by 6%, and has maximum flexibility through the individual tunability of the three key bump function parameters. Another feature is that multiple bump circuits can be easily cascaded to represent multivariate probability.A vision of the futureWith power scaling in CMOS tapering off, there has been renewed interest in analogue computation recently, and the researchers expect to see some very exciting results in this area. They are currently working to integrate low-power circuits, such as their bump circuit, into larger systems for real-world applications."One application area we've been working on is machine vision," said Lu. "We've been working with image processing and machine vision researchers to build a complete pipeline using analogue circuits. This circuit helps to provide a path to implementing multi-dimensional kernel methods for machine learning."Systems using the bump circuit could find application in many areas such as healthcare monitoring, environmental monitoring, process control and battlefield surveillance. In addition, the nano-power tunable linear transconductor developed in this work, which has the advantages of ultra-low power, large input range and gm tunability, could be used in a huge range of applications such as amplifiers, filters and oscillators.Related products:LMV1031UR-20LM4889MALM4867MTE

Parts fail and things break. It's a fact of life and engineering. Some component failures can be avoided by good design practices, but many are out of the hands of designers. Identifying the offending component and why is might have failed is the first step to refining the design and increasing the reliability of a system that has been experiencing component failures.How Components FailThere are numerous reasons for why components fail.Some failures are slow and graceful where there is time to identify the component and replace it before it fails completely and the equipment is down. Other failures are rapid, violent, and unexpected, all of which are tested for during product certification testing. Some of the most common reasons for components to fail include:Over currentOver voltageOver temperatureConnected incorrectlyChange in operating environmentManufacturing defectMechanical shockMechanical stressRadiationContaminationPackagingConnectionsAgingCascading failureCorrosionRustingOxidizingThermal runawayLoose connectionsElectroStatic Discharge (ESD)Electrical stressBad circuit design Component failures do follow a trend. In the early life of an electronic system, component failures are more common and the chance of failure drops as they are used. The reason for the drop in failure rates is that the components that have packaging, soldering, and manufacturing defects often fail within minutes or hours of first using the device. This is why many manufacturers include a several hour burn in period for their products.This simple test eliminates the chance a bad component can slip through the manufacturing process and result in a broken device within hours of the end user first using it.After the initial burn in period, component failures typically bottom out and happen randomly. As components are used or even just sit, they age.Chemical reactions reduce the quality of the packaging, wires, and the component, and mechanical and thermal cycling take their toll on the mechanical strength of the component. These factors cause failure rates to continuously increase as a product ages. This is why failures are often classified by either their root cause or by when the failed in the life of the component.Identifying a Failed ComponentWhen a component fails there are a few indicators that can help identify the component that failed and aid in troubleshooting electronics. These indicators are:Visible-The most obvious indicator that a specific component has failed is through a visual inspection. Failed components often have burnt or melted areas, or have bulged out and expanded. Capacitors are often found bulged out, especially electrolytic capacitors around their metal tops. IC packages often have a small hole burned in them where the hot stop on the component vaporized the plastic around the hot spot all the way through the IC package.Smell- When components fail, a thermal overload often occurs which causes the magic blue smoke and other colorful smoke to be released by the offending component. The smoke also has a very distinct smell and varies by type of component. This is often the first sign of a component failure beyond the device not working. Often the distinct smell of a failed component will stay around the component for days or weeks which can aid in identifying the offending component during troubleshooting.Sound- Sometimes components make a sound when they fail. This happens more often with rapid thermal failures, over voltages, and over current events. When a component fails this violently, a smell often accompanies the failure. Hearing a component fail is rarer, and it often means that pieces of the component will be found loose in the product so identifying the component that failed may come down to finding which component is no longer on the PCB or in the system.Testing- Sometimes the only way to identify a component that has failed is to test individual components. This can be very challenging on a PCB since often other components will influence the measurement since all measurements involve applying a small voltage or current, the circuit will respond to it and readings can be thrown off. If a system uses several subassemblies, often replacing subassemblies is a great way to narrow down on where the issue with the system is located. 

Inside a secretive AI nonprofit backed by Elon Musk and other Silicon Valley figures, a handful of robots designed to help out in warehouses are gradually learning how to do useful household chores.OpenAI, which was created to do basic AI research, is reprogramming robots developed by Fetch Robotics, a company that supplies warehouse automation hardware. Researchers at OpenAI are equipping the robots with software that lets them train themselves through trial and error.The effort reflects a bet that innovations in software and machine learning, rather than breakthroughs in hardware, are the way to give robotics remarkable new capabilities. Fetch makes a range of robots for warehouses, including systems that follow workers around a building, carrying items dropped into a basket. OpenAI is using a system that features a mobile base but also 3-D depth sensors, a 2-D laser scanner, and a robotic arm with seven degrees of freedom.In April, OpenAI recruited Pieter Abbeel, a professor at the University of California, Berkeley, and a leading expert on robot learning. Abbeel has shown how robots can use a machine-learning approach called deep reinforcement learning to acquire completely new skills that would be hard to program by hand, such as folding towels or retrieving items from a refrigerator. Google DeepMind, an AI subsidiary based in the U.K., uses this technique to get computers to play computer games at a superhuman level (see “Google’s AI Masters Space Invaders”).Abbeel’s robots learn tasks from scratch, using a neural network that receives sensor input and controls physical movement. The network adjusts its parameters automatically as it inches closer to its goal. A robot might try thousands of grips, for instance, in the process of learning how to hold a certain object.“If this goal can be achieved, then there will be economic and industrial benefits,” says Marc Deisenroth, an expert on reinforcement learning at Imperial College London. “Imagine a Roomba not only cleaning your floor but also doing the dishes, ironing the shirts, cleaning the windows, preparing breakfast.”Deisenroth says using off-the-shelf robots could drive costs down. “Currently, the software seems to be the bottleneck,” he adds. “However, independent of this, better hardware could also lead to substantial improvements.” Soft manipulators and elastic feet similar to a monkey’s feet are concepts that researchers have started working on, he says.Some manufacturers, including the Japanese company Fanuc, are testing reinforcement learning as a way to train industrial robots quickly in new tasks such as learning to grasp unfamiliar objects. When many robots work in parallel, the training time required is reduced accordingly. Robot researchers at Google are testing similar learning techniques.“Moving away from having to program robots by hand by endowing robots to learn autonomously is a key element for the future of robotics,” says Jens Kober, an expert on robot learning at Delft University of Technology in the Netherlands. Kober says having robots share the information they have learned will be crucial.While robots such as those made by Fetch are finding their way into many factories and warehouses, domestic robot helpers remain the stuff of science fiction. Performing seemingly simple tasks like washing dishes or folding laundry in a messy home setting is incredibly hard for a machine. A robot programmed the conventional way can easily be thrown off by an unfamiliar object or a slight variation in lighting.OpenAI confirmed that it is working with the robots from Fetch, but it declined to comment further. Melonee Wise, the company’s founder, couldn’t be reached for comment (see “Innovators Under 35: Melonee Wise”).OpenAI was created by Musk and a handful of well-known (and well-heeled) Silicon Valley entrepreneurs, including investor Peter Thiel, Y Combinator president Sam Altman, and the incubator’s cofounder Jessica Livingston. The nonprofit’s backers have committed $1 billion in funding to the project, and it is being led by Ilya Sutskever, a prominent AI researcher who left Google to join the project, and Greg Brockman, an early employee at the high-profile digital payment company Stripe.While OpenAI has committed to making the technology it develops publicly available, it could certainly benefit companies backed by Musk and Thiel, as well as those emerging from Y Combinator. 

The electronics world has been dreaming for half a century of the day you can roll a TV up in a tube. Last year, Samsung even unveiled a smartphone with a curved screen—but it was solid, not flexible; the technology just hasn't caught up yet.But scientists got one step closer last month when researchers at the U.S. Department of Energy's Argonne National Laboratory reported the creation of the world's thinnest flexible, see-through 2-D thin film transistors.These transistors are just 10 atomic layers thick—that's about how much your fingernails grow per second.Transistors are the basis of nearly all electronics. Their two settings—on or off—dictate the 1s and 0s of computer binary language. Thin film transistors are a particular subset of these that are typically used in screens and displays. Virtually all flat-screen TVs and smartphones are made up of thin film transistors today; they form the basis of both LEDs and LCDs (liquid crystal displays)."This could make a transparent, nearly invisible screen," said Andreas Roelofs, a coauthor on the paper and interim director of Argonne's Center for Nanoscale Materials. "Imagine a normal window that doubles as a screen whenever you turn it on, for example."To measure how good a transistor is, you measure its on-off ratio—how completely can it turn off the current?—and a property called "field effect carrier mobility," which measures how quickly electrons can move through the material."We were pleased to find that the on/off ratio is just as good as current commercial thin-film transistors," said Argonne postdoctoral scientist and first author Saptarshi Das, "but the mobility is a hundred times better than what's on the market today."The team also tried bending the films to test what happens under stress. In most thin film transistors, the material starts to crack, which, as you might imagine, affects performance. "But in ours, the properties didn't change at all," Roelofs said. "The layers just slide and don't crack."The transistors also maintained performance over a wide range of temperatures (from -320°F to 250°F), a useful property in electronics, which can run very hot.To build the transistors, the team started with a trick that earned its original University of Manchester inventors the Nobel Prize: using a strip of scotch tape to peel off a sheet of tungsten diselenide just atoms thick."We chose tungsten diselenide because it provides the electron and hole conduction necessary for making transistors with logic gates and other p-n junction devices," said Argonne scientist and coauthor Anirudha Sumant.Then they used chemical deposition to grow sheets of other materials on top to build the transistor layer by layer. The final product is 10 atomic layers thick. (See sidebar for an illustration).Next, the team is interested in adding logic and memory to flexible films, so you could make not just a screen but an entire flexible and transparent TV or computer."However, more work needs to be done in developing large-area synthesis of tungsten selenide to realize the true potential for applications of our work," said Sumant.    

VTT Technical Research Centre of Finland has developed a method for the manufacture of thin film transistors using a roll-to-roll technique only. Thin film transistors can now be manufactured using roll-to-roll techniques, such as printing, for the deposition of patterns on the substrate layer of film. This is set to expand the range of electronic components and products, while slashing their production costs. Thin film transistors are more suitable than traditional silicon chip transistors for applications such as large-surface display screens, certain sensor applications, toys, games and smart cards.A transistor is a basic electronic component which can function as an electrical switch, an amplifier or a memory element. For transistor technology, roll-to-roll fabrication techniques have a range of advantages. These include the possibility to use large surface areas, as well as mechanical flexibility, transparency and low production start-up costs. Until now, production of thin film transistors has typically been only partly based on roll-to-roll techniques, resulting in fairly high mass production costs.As the technology matures, it is predicted that the markets for thin film transistors will grow from their current value of three million dollars to around 180 million over the next decade.VTT has developed thin film transistor production techniques as part of the EU POLARIC research project. With the aid of a special self-aligning technique, the method under development eliminates the challenge of aligning the patterns in the different thin film layers accurately against each other in the roll-to-roll process. In addition, the pattern size for transistor components is pushed to the limit of minuteness possible for printing techniques; this means patterns of a few dozen micrometres at their tiniest..Producing thin film transistors using a self-aligning roll-to-roll manufacturing process is one of the few demonstrations internationally so far. Initial experiences of this thin film transistor manufacturing process are promising. It provides VTT with the ideal basis for using the process to test thin film materials as they develop, to develop more complex electronic circuits and to trial various applications. The goal is to keep developing the technology until it matures enough to provide a springboard for new business activities. VTT is now seeking companies interested in developing applications based on printed thin film transistors. 

Texas Instruments today announced the industry's first flexible high frequency 13.56 MHz sensor transponder family. The highly integrated ultra-low-power RF430FRL153HCRGER system-on-chip (SoC) family combines an ISO 15693-compliant Near Field Communication (NFC) interface with a programmable microcontroller (MCU), non-volatile FRAM, an analog-to-digital converter (ADC) and SPI or I2C interface. The dual-interface RF430FRL153HCRGER NFC sensor transponder is optimized for use in fully passive (battery-less) or semi-active modes to achieve extended battery life in a wide range of consumer wearables, industrial, medical and asset tracking applications.Non-volatile FRAM combines the speed, flexibility and endurance of SRAM with the stability and reliability of flash – while providing the industry's lowest power consumption and virtually unlimited write cycles. FRAM allows developers to create products that can quickly store sensor data and enables easy configuration of the transponder and sensors to meet any application's needs.Integrating NFC sensors into medical, industrial and asset-tracking applicationsDevelopers can now design products that require an analog or digital interface, data-logging capabilities and data transfers to an NFC-enabled reader. The RF430FRL153HCRGER transponder acts as a sensor node for these applications and generates an IoT-ready solution when an NFC-enabled device pushes the data to the cloud.In medical or health and fitness applications, the RF430FRL153HCRGER can be used in disposable patches that sense temperature, hydration and more. This allows patients to monitor and share vital data securely with their health providers. The device monitors and logs data in local storage (FRAM) before transferring it to an NFC-enabled tablet or smartphone.The RF430FRL153HCRGERenables the design of maintenance-free and hermetically sealed galvanic isolated sensor systems in the industrial markets. These sensors are powered out of the RF field and communicate wirelessly through NFC to collect and log data.Logistics applications such as food tracking need constant temperature control, which can be monitored and logged with the RF430FRL153HCRGER transponder. It allows the design of highly integrated, size-optimized and easy-to-use data loggers with several sensors that connect to NFC-enabled devices and readers throughout the distribution channel.Features and benefits of TI's RF430FRL153HCRGER NFC sensor transpondersSupports wireless communication via the ISO/IEC 15693, ISO/IEC 18000-3 compliant RFID interface.Optimized for 1.5 V single-cell-battery-powered designs or battery-less designs that harvest energy from the RF field generated from an NFC reader at the same reading distance. Intelligent power management includes a battery switch to ensure long battery life.14-bit sigma-delta ADC with ultra-low input current, low noise and ultra-low offset enables developers to connect up to three additional external sensors in addition to the integrated temperature sensor.SPI or I2C interface can support digital sensors or connect the device to a host system.Application code embedded in ROM manages RF communication and sensor readings to provide the ultimate flexibility in configuring the device. Developers can configure sampling rates, measurement thresholds and alarms.Universal non-volatile memory (FRAM) allows data storage as well as extension and adjustment of application code.Integrates a 16-bit ultra-low-power programmable MSP430 CPU core that is supported by a robust ecosystem of development tools.Fully integrated into TI's Code Composer Studio (CCS) and IAR's Embedded Workbench® integrated development environments (IDEs).