US Start-Up Takes a Software-Centric Slant
US start-up Helium launched into the Internet of things this week, with a refreshingly different approach.
Helium provides sensors that are simple to configure and connect over a low-speed, low-power wireless network to gateways passing data to its cloud service. The cloud facility offers device management, data storage, analytics, machine learning and a software update service. Helium is aiming at the industrial and enterprise Internet of things markets, with a focus on retrofitting to existing machinery.
Temperature sensing is an initial key area, and the company will provide sensors for various devices including the likes of medical refrigerators that store blood, medicines and vaccines. Such machines require a high degree of monitoring and regulatory compliance, and much of this is currently carried out manually on paper systems. Helium also has sensors for metrics including humidity, light, noise, motion and pressure.
The company is offering a full system that’s heavily focused not only on ease of use, but also on the simplicity of configuration with existing machinery not designed for the Internet of things.
Helium’s approach is pleasingly different. Most of today’s Internet of things systems are hardware-centric, to minimise costs. The software consequently lives in read-only memory, so the sensor can’t be reprogrammed easily and processing at the edge is minimised. This means that all data is passed to a gateway, which may carry out local processing or pass the information back into the cloud. The overall result is a lack of programmability.
Helium’s system is software-centric. The functionality — and price — of the sensors themselves is higher because they have more processing power and memory, but the goal is to reduce the whole-life cost of the system by making it flexible and reprogrammable. There are several parts to achieving this.
First, there is the opportunity to carry out “smart sensing”, in which the sensor only sends data when it needs to do so. For example, it may not be important to send information frequently or at all during normal operating conditions. Users will need to learn how to optimise this based on the data the system provides, but this allows for savings in sensor battery life, maintenance visits, data transmission, data storage, processing and analytics.
Secondly, Helium’s approach means that the smartness of the entire system can be updated with new rules or exceptions as part of a software update. This could be useful when incorporating technical changes or operating differences across industries. It could also be beneficial should compliance requirements change, preventing the need for replacement of all the sensors in a system.
Thirdly, large numbers of Internet of things players are launching new hardware, software and cloud solutions. The industry is in an early phase of its development, and a high degree of change can be expected as companies succeed or fail, as hardware develops and software architectures adapt, as a wider array of cloud services become available and as analytics develop. For users it will be prudent to design an Internet of things system on the assumption that it will need many software updates during its lifetime.
There will be strong implications if the software-centric model comes to rule the Internet of things. Although there will be constant focus on the reduction of hardware costs, it’s likely that Internet of things will see the software-drives-computing spiral seen in other markets — notably for PCs and, recently, mobile phones — to some extent.
It’s unclear whether it’s the right time to start pushing a different architectural vision, when the Internet of things is in trial mode for many industries and enterprises, and when whole-life costing and business cases for the Internet of things aren’t well understood. Customers may not yet feel competent enough to make the trade-offs between less-expensive sensors and system flexibility.
The industry will need to develop machine learning as a highly distributed function, with aspects carried out in all parts of the network, including at the edge. However, this would require machine learning to no longer be viewed as something that needs the full scale and power of centralised cloud services in order to be economic and useful. Chipset makers including MediaTek and Qualcomm are already starting to build various machine learning functions into smartphone components, and we expect to see a great deal of development in this area.
These are potential considerations for the whole industry. Helium is right, as a start-up, not to focus too heavily on the vision, but instead to identify key ways that its system can enable rapid early adoption and demonstrate the benefits of its architecture.
|
If you’d like to receive free Daily Insight |