Horizontal Technology Landscape of IoT Stack — Sensor to Cloud
By Somshubhro Pal Choudhury
In the previous article we covered IoT innovations in Applications and Market verticals, a survey of 500+ global application focused IoT startups. We also covered a broad landscape of the Indian IoT Ecosystem that we are tracking at IoTForum (an initiative by TiE Bangalore and IESA) with over 600+ IoT Startups.
As each technology component can go down to any level of depths, my intention here is to give a high-level representation of the market and technology trends primarily using visuals.
More than Moore…
It is well-established fact that Moore’s law has slowed. Even though we are in 7 nm technology, the pace of change has slowed down primarily because of the cost of moving to a new technology node, the challenges it takes to adopt and also to achieve ROI on the previous node. The effort is now more lateral integration, System-in-a-Package (SIP), flexible/printable electronics, ultra-low power, RF and sensor integration, printable/flexible electronics and so on, commonly referred to as “More than Moore”. As we shift to more IoT devices, the need is not performance unlike the Mobile, Servers, and PCs, but it is ultra-low power, form factor, more forward and backward integration of the signal chain, innovative packaging technologies and lowering cost.
Way forward for Sensors
There are three market verticals where sensors have played an enormous role; prices are down significantly leading to mass adoption. These markets are Smart Phones, Wearables, and Automotive.
While incremental cost and data accuracy improvement continue to happen in the above sensor technologies, there is a big focus on cost reduction of these sensors with China taking the lead and more innovation on sensor algorithms and sensor fusion based on application level innovation primarily in wearables and smartphones. Another clear direction is integration, for example, MEMS sensors with several degrees of freedom integrated with processor core in a single package running algorithms like Kalman Filtering.
A quick survey of the sensor startups showed that the innovation in the sensor world is more focused on the industrial and commercial use cases that include innovative hardware and as well as algorithms to make better sense of the data using Machine Learning or pure play fundamental equations. The scope of innovation is ripe in these sectors as the technology sometimes used is age old, with high cost and hence perfect for disruption.
Innovative sensors are being developed for industrial use cases, spectral imaging for food quality, cheaper and more effective water & air quality sensors, hazardous chemical detection, nano sensors, micro and nano actuators, biometric sensors, sensor patches & tattoos based on flexible/printable electronics and odor sensors. Startups are doing miniaturized spectroscopy, ultrasound acoustic sensor, micro-radar on a chip, below the skin surface layer fingerprint sensing, optical MEMS actuators, corrosion detection, 30X more sensitive CMOS image sensor and so on…
The energy harvesting space is still early with a lot of ongoing research. It can be broadly classified in different categories as highlighted in the figure. Wireless Charging is now a reality and improvements are coming with electromagnetic radiation based energy harvesting that can charge at larger distances. Deriving energy from ambient light is a reality today.
While for IoT Devices energy harvesting is the eventual way to go, I recently did some work on the landscape for energy storage and hence including it out here. While Fuel Cells are the Holy Grail, it still is not for the large volume mass market. While incremental improvements in Lithium-Metal batteries is ongoing, last few years have seen renewed interest in new non-Lithium battery chemistries but they are in a long and arduous commercialization process. Super/ultra Capacitors are finding interesting use cases.
With IoT and AI/Deep Learning, there is a resurgence of Processors, co-Processors, and accelerators. The effort is both on the cloud and the edge with AI/Deep learning accelerators and co-processors, edge analytics to do more intelligence at the edge and efforts on Opensource RISC-V processor.
Development efforts on AI Co-Processor include Google Tensorflow TPU, IBM TrueNorth, startups like KnuEdge (raised $100M), UK based Graphcore (raised $30M) with NVidia already a formidable player in this field. FPGA based Accelerators are being used by Microsoft and Tencent for AI/Deep learning acceleration, Intel has acquired Nervana and Altera and we should see them in some shape and form working together with the server CPUs. DeepPhi has been acquired by Xilinx.
The effort on the edge is quite diverse with ARM, Intel, Samsung and Qualcomm leading the fray with ARM Dynamic IQ, Intel’s acquisition of Movidius (just launched a USB dongle based AI capability) and MobileEye, Samsung M1 Architecture, Qualcomm Zeroth Neural and many startups — EyeRiss from MIT (claiming 10x faster than GPU), Brainchip, TensTorrent and Mythic. The RISC-V open source Instruction Set Architecture (ISA) effort was started at the University of Berkeley in 2010. Now several companies are embracing RISC-V including large semiconductor companies and as well as startups.
Multiple wireless technologies are at play depending upon the IoT context with Bluetooth, Wi-Fi, Zigbee, ZWave, 6LowPan, Wireless-HART all having specific niches in home, consumer, buildings or industrial spaces.
But the fight is on Low Power Wireless LAN (LPWAN) technology where multiple standard and non-standard based technologies are in the fray. While SigFox and LORA had an early start, NB-IOT and LTE-M are the emerging standards for IoT LPWAN. Seems like the industry is divided with large established operators cosying up to NB-IoT in the licensed spectrum and already running several trials. The Tier 2/Tier 3 operators and new entrants in many countries have embraced LORA and it has a larger fan following because of its unlicensed spectrum use. In India, for e.g Tata Communications who has been in the backhaul and wired enterprise communications space has had an early start with LORA while Reliance, Vodafone, and Airtel are busy with the 4G war.
IoT Gateway Evolution
Till very recently, primarily because of Consumer IoT and Telematics (the early movers of IoT), all the data was being sent back to the Cloud for storage, processing, and analytics.
With Industrial IoT, the intelligence is being pushed to the gateway. Many industrial enterprises do not want to send the data to the cloud sighting issues including security, cost, latency and simply non-availability of a network like in remote oil & gas situations. Multi-protocol Software Defined Networking (SDN) based gateways which can translate multiple protocols both North and South of the IoT stack, deployment of more servers at the gateway or aggregation point for storing and analyzing sensor data and a scaled down version of the Cloud platform along with analytics are the trends. Various standardization efforts are on with Fog Consortium and Rooftop Computing, while companies like Foghorn and even Microsoft Azure now has an edge version.
Experts will debate on the best way to classify IoT platforms, not all platforms are the same, widely differing in functionality. But broadly speaking they are targeting either towards developers across multiple verticals (e.g Azure) or end-user focused towards specific verticals (e.g Jasper or Gobee). The latter implies a narrower focus on specific verticals, less of coding and more of application development with better richer AI and interfaces with less software effort.
From a revenue standpoint, AWS, Azure does not quote numbers for IoT specifically, but the top three today seem to be ThingWorx, Jasper and C3 IoT. A bunch of them are in the $1–10M revenue range. Most platform companies are either building more verticalized solution focused on specific verticals or sub-verticals, building more specialized features like analytics and AI while others are simply focused on building up the customer base as rapidly they can. This category has seen the largest number of M&As so far in IoT category. We should expect to see clear winners, rapid consolidation, and death of plenty of these platform startups very soon.
Domain specialization in IoT Analytics
Mostly streaming of IoT data would require handling by data streaming engines as real time or near real-time cannot afford the latency of conventional batch processing. IoT data is also very time and location dependent — spatial and timing co-relation exists which can be taken advantage off.
While several analytics platforms and frameworks exist, both open source and by the big players, there are several players building a service or a layer on top of the basic analytics frameworks. Similar story as the IoT Platforms, these platforms are targeted either towards developers or end users. The so called analytics platform provides an additional layer which eases the development of domain specific applications. A marketplace model for algorithms is also emerging with Google investing in Algorithmia.
Today most IoT platform companies are pulling in the analytics frameworks inside as part of their evolution and specialization or at least providing the interfaces to it. But the need and challenge is on domain specific analytics for each vertical or sub-vertical. Even though several companies are playing in this field, in the supply chain, predictive analytics etc., space is wide open and winners will likely be the vertical/sub-vertical oriented domain specialists who have paired up with analytics experts writing their analytics engines on opensource or as an App on top of the larger proprietary platforms. Oil & Gas and aircraft engine maintenance already with humongous amounts of data are leading in using domain specific analytics.
While several analytics platforms and frameworks exists, both opensource and by the big players, there are several players building a service or a layer on top of the basic analytics frameworks. Similar story as the IoT Platforms, these platforms are targeted either towards developers or end users. The so called analytics platform provide an additional layer which eases the development of domain specific applications. A marketplace model for algorithms is also emerging with Google investing in Algorithmia.
Today most IoT platform companies are pulling in the analytics frameworks inside as part of their evolution and specialization or atleast providing the interfaces to it. But the need and challenge is on domain specific analytics for each vertical or sub-vertical. Even though several companies are playing in this field, in supply chain, predictive analytics etc., the space is wide open and winners will likely be the vertical/sub-vertical oriented domain specialists who have paired up with analytics experts writing their analytics engines on opensource or as a App on top of the larger proprietary platforms. Oil & Gas and aircraft engine maintenance already with humongous amounts of data are leading in using domain specific analytics.
The focus is also shifting from rules engines and visualization to more predictive and prescriptive analysis, to derive more insight and foresight, where again the domain knowledge is a must.
Visial analytics, both machine vision and ML based seem to be the hottest topic os the startups today followed by speech.
IoT Security is of prime concern today for the adoption of IoT. Stories abound on how video cameras, door locks, and cars have been hijacked and created mayhem in security. See some of the largest breaches in this excellent interactive infographic.
Today’s IoT Security is based on the Internet Age and not capable for the IoT age. From a IoT perspective, it is complex, costly and vulnerable. You break one edge device and you hack them all. Most devices are not secured running on default passwords and keys and are not fail safe. Practically, it is impossible to secure and maintain security for billions of devices made by ODMs and EMS companies who operate today based on margins off a one-time sale and has no incentive or know-how to keep the devices secured well into the future. Even if there is a mandate imposing product liability for IoT breaches, the economics does not work out to maintain and upgrade the security of billions of devices by the OEMS/ODMs. The challenge is to address the identity of these edge nodes, rights management, maintaining data integrity and privacy for the IoT data. Any security needs to run in a severe resource constrained environments with less processing power and memory footprint.
A recent article by Arvind Tiwary who heads IoTForum in India titled IoT Security: Lessons from Physical world details the practicality of the physical world security that needs to be practiced for IoT Security. The recent IoT security breaches reinforce the issue that we are losing the battle and maybe have the wrong approach to handling IoT Security. Indian government launched an ambitious scheme to monitor internet traffic real time and help isolate restrict rogue elements.
IoT Security investments are at an all time high. The following are the broad areas where innovation is happening in IoT Security
- Tools and services that evaluate the software at edge and gateway nodes, scan for vulnerabilities and fixing them Microsec, Argus Security etc.
- Identity Management of IoT edge nodes via Physically Un-clonable Function(PUF), and managing changing/rotating security keys (Device Authority).
- AI/ML analysis for anomaly detection of threats and suspicious behavior like Countertack. Isolate and restrict suspicious and corrupt nodes from the Gateway. Stay current on all security threats connected to the cloud but pushed down to the gateway. Examples include startups like Bayshore Networks, CyberX etc. A pretty interesting approach on security is provided by a startup PFP Security that monitors detailed power profile of each application to determine changes in behavior patterns for threat analysis.
- Complete end to end sensor to the cloud security using some or all of the above techniques like Mocana.
Vertical market efforts on security are focused on primarily the connected car (Argus Security, Karamba Security), industrial (Indegy) and home/consumer (Dojo Labs).
About The Author:
About Bharat Innovations Fund:
Bharat Innovations Fund was created to encourage and support bright entrepreneurs with the potential to create disruptive innovations that can solve some of India’s toughest problems. Twitter: @bharat_fund