CEVA is a global leader that provides the semiconductor industry with digital signal processing (DSP) silicon intellectual property (SIP). The firm has been developing DSP technology since 1991. DSP is capable of real-time data processing and can satisfy and support applications that require real-time response without any delays.
Some of DSP’s mainstream applications are ultra-high-speed voice processing, background noise elimination, data compression, and signal quality and efficiency enhancements that require mathematical algorithms.
CEVA’s DSP cores can help smart devices offer clearer sounds, sharper images and faster data processing, and various smart devices such as smartphones and wearable devices adopt DSP technology.
CEVA is listed on NASDAQ and more than 7.5 billion chips carrying CEVA technology have been widely used in telecommunication, network, multimedia and Internet of Things (IoT) applications. During the CEVA Technology Symposium 2016, Mr. Moshe Sheier, director of strategic marketing at CEVA, gave an interview where he talked about the trends and future developments of DSP applications.
CEVA targeting the rapid growth of cellular IoT devices
The semiconductor industry has seen a lot of major mergers in recent years to cope with market and technology process changes. For example, Softbank acquired ARM for an astonishing figure and it has been one of the most significant acquisitions in the semiconductor industry in 2016. Now many firms have made significant moves and investments in IoT, racing to establish a foothold in what is seen as the most promising emerging sector in the IT industry. With the smartphone and tablet markets close to saturation, the next big thing will be IoT applications.
Many IoT devices have relied on smartphones that support short-distance communication protocols, such as Bluetooth and Zigbee, which has limited the growth of IoT’s popularity. And industrial IoT applications face different kind of problems when used outdoors or in the wilderness, such as a lack of electric sockets or WiFi access points. With the coming of Industry 4.0, IoT devices have entered a new era where low power consumption, low cost and long-distance connection are vital. The connectivity problems crippling outdoor IoT devices are one of the crucial problems that need to be solved.
International telecommunication associations and organizations like 3GPP have sped up the standardization of NarrowBand IoT (NB-IoT), allowing firms in the industry to step up efforts to develop various NB-IoT nodes for different market segments. Because of this, Sheier sees a new generation of IoT devices moving towards being independent of smartphones by relying on more sensors, edge processing and low power wide area technologies. As telecom firms are looking to use LTE frequencies to construct NB-IoT systems and adopt LTE-compatible infrastructures, many major chip vendors have devoted large amounts of resources to developments for cellular IoT networks. Now the industry is eagerly anticipating IoT devices with multiple sensors and long-distance wide area connectivity.
According to market research done by Ericsson, the number of long-distance cellular IoT devices will likely grow from 400 million units in 2016 to 2.1 billion by 2022 with a 30% cumulative annual growth rate (CAGR). The market research also predicts that by 2018, the total number of IoT devices will exceed the number of mobile phones. In terms of semiconductor chips, smartphones account for 1-2 billion devices per year but chip demand from IoT devices may soar to 30-50 billion cumulative units, driving powerful growth in the semiconductor market. Sheier is very optimistic about the future of cellular IoT.
Computer vision and image recognition DSP applications lead machine learning boom
Computer vision and image recognition technologies used in self-driving vehicles and drones are an important development for DSP applications. In addition to improving the quality of the images, CEVA has invested a great amount of resources to develop vision applications, and with always-on, always-sensing, and always-connected technologies, combined with Big Data processing and artificial intelligence (AI) technologies for machine learning, CEVA’s DSP has successfully expanded its reach from monitoring and security systems to emerging applications such as smart city and smart medical ones. The technology has also been welcomed by mainstream brand vendors.
CEVA works with strategic partners on developing edge AI technologies and uses neural network to develop machine learning applications. This year, CEVA has introduced the second generation of CEVA Deep Neural Network (CDNN2) software framework to further simplify the machine learning process for low power embedded systems.
Combining lower power consumption and machine learning technologies, CEVA’s strategic partner, emza Visual Sense, showcased an IoT visual sensor with power consumption of only 2mW at this year’s symposium. With support by always-on technology, a DSP and special software algorithm, it allows IoT devices to perform automatic image recognition. The visual sensor can be operated over a long period of time using batteries and is a highly flexible solution.
This type of battery-powered AI visual sensors that can be used for long periods of time is quite suitable for long-term and elderly care. Although such devices may not offer high-resolution images or replace monitoring systems equipped with high-end cameras, they offer smart medical and smart city solutions featuring machine learning and real-time response.
Software integration is the key
The development of Big Data analytics and AI has been stimulating the development of different IoT systems. Sheier cited examples from the pharmaceutical and biomedical sectors. During the transportation of pharmaceutical products, temperature control is crucial. The data collected through various sensors of the IoT systems and processed using different software algorithms can predict the impact on the pharmaceutical products caused by different transportation vehicles and the length of transportation time. Based on the information, remedial measures can be adopted to make sure the products arrive in time for the patients. This example shows the important role that software with precise algorithms play.
With the increasing popularity of machine learning and Big Data analytics, a lot of software now comes with value-added services. To accommodate this trend, CEVA has been focusing on enhancing the development environment and interface for software, in particular, the machine learning technology based on the deep neural network. Addressing embedded systems’ limitations in memory and computing abilities, CEVA has developed offline operation environment by setting the development of machine learning in exterior systems and using training framework and library such as Caffe and TensorFlow to transfer the result of machine learning onto the CEVA embedded processors.
Using CEVA Network Generator, this type of software development for machine learning systems can shift the complicated neural network structure and weight, and turn it into a customized neural network enabling real-time response for embedded systems that are undermined by power loss and lack of memory. It is meant for IoT applications that require fast response time, have low tolerance of delay, and do not wish to rely on cloud network. With CEVA’s new CDNN2 structure and the highly flexible and customizable assessment and deployment kit (ADK), the integration with more sensors and smart devices can become easier.
CEVA Director of Strategic Marketing, Moshe Shei
Based on many years of development and successful experiences in the IP industry, CEVA has won strong recognition for its innovative skills on quality and technology from Taiwan-based firms. The easy access to DSP and connectivity IP allow Taiwan-based IC design houses and OEM/ODM firms to target different market segments and avoid price competition by developing differentiating products for emerging applications such as cellular IoT. This is an opportunity to ride the IoT wave and achieve a win-win situation.