Part of the  

Chip Design Magazine


About  |  Contact



IoT vs. Traditional Embedded for Analog, Low Power and Security

By John Blyler, Chief Content Officer

In Part III, technology leaders from STMicro, Atmel, Mouser, Synopsys, Movea, and ARM, define the big challenges in IoT – mixed signal, low power and security.

Will the Internet-of-Things (IoT) bring new analog integration, low power and security challenges to traditional SoC and embedded designs? To answer these questions, System Design Engineering sat down Bernard Kasser, Director Security R&D, Advanced System Technology, STMicroelectronic; Bob Martin, Senior Manager Microcontroller Group Central Applications, Atmel; Kevin Parmenter, Director of Technical Resources, Mouser; Steve Smith, Senior Director of Marketing, AMS Group, Synopsys; Cyrille Soubeyrat, VP of Engineering at Movea; and Diya Soubra, CPU Product Marketing Manager, ARM. What follows are excerpts from that discussion. – JB

Blyler: Analog mixed signal integration with MCUs is a prerequisite for the IoT. How difficult is it for designers – especially digital designers – to incorporate analog and mixed signals into their SoCs? What tools are available? What are the best practices?

Kasser: The vast majority of IoT devices today are powered by MCUs running at modest frequencies unlike application-processor power-houses at the heart of mobile devices. But the IoT MCUs present formidable integration challenges as they come packed with all sorts of I/O and signal-acquisition peripherals. The deployment on the same silicon of traditional digital logic and datapaths can easily be overwhelmed by the typical mixed-signal MCU SoC backend design process, which can be complicated by issues such as placement of analog macros and synthesized logic, by timing-closure accounting, by poorly characterized analog IPs, and by signal integrity of digital signals and cross-talk introducing noise for analog blocks (think of ADC, DACs, bandgaps, etc.). So designers must carefully simulate their mixed signal designs to insure the behavior of control logic and analog circuits working together. These simulations can take days and weeks and, may still require substantial handcrafted black magic from practitioners of the (analog) art.

Martin: Mixed signal design has been done for years and the IoT focused products for the most part does not require cutting edge analog integration.  The type of things being monitored in a household for example are relatively slow and easily conditioned to be interfaced to the microcontrollers being used to implement the ‘motes’ or nodes. The IoT factor is the extension of taking existing microcontroller and analog front end (AFE) and making this data available to the could and subsequently allowing the cloud to control these nodes.  The real challenge for the chip designers is how to best balance feature sets against power consumption as many of these endpoint applications become very application focused.   It would seem that the shift from the MHz race to the milli-watt race over the past few years has left the EDA tools manufacturers playing catch-up to some extent.  Additionally accurate characterization of the current chip process technology in the static state (non clocked) is now much more critical since most of the IoT nodes spend most of their time asleep.

Parmenter: Essentially open source hardware and reference designs from the semiconductor companies have lowered the barriers to entry substantially. We are seeing some interesting open source hardware, specifically the BeagleBone Black from TI and others. Reference designs are becoming more common. In many cases much of the work has been done by the semiconductor companies.  It’s important to understand system level interfacing because you still have to read analog signals and keep noise out of the system. Your power supply has to be properly designed and stable. More designs have wired and wireless communications interfaces so the make versus buy decision is always a moving target. So it comes down to clever software design and selection of an OS and other software after the decision to make versus buy hardware decisions are made.

Smith: The IoT generally assumes very compact, low-power, connected devices that can be applied to everyday living applications. Thus, the assumption is that they include one or more processors – these can be discrete, using traditional or embedded microcontrollers (MCUs). Additionally, these devices require connectivity – this could be done locally via near-field communication (NFC), Bluetooth, or WiFi. The wireless components will likely be discrete due to cost, time-to-market, and multi-sourcing requirements. Likewise, any required sensors, such as those for motion detection, geolocation, temperature, and so on, will be highly integrated in themselves but still discrete. In this case, the integration is done with appropriate packaging technologies such as package on package (PoP), multi-die packages, etc.

Thus, this is mostly a system-level challenge – how to create models for each of the components and how to develop the software to drive and manage the devices. However, where there is a need to integrate at the chip level, it’s important that the entire SoC can be designed in an efficient manner. The co-design of digital and analog parts must be available and understood by engineers who are responsible for pulling together the entire chip.

Soubra: Today, mixed signal design is totally different than that of ten years ago thanks to EDA tools that have made a radical change to the design process. These tools give you views into both the digital and analogue side. For example, you can trace through the assembly code to see the digital signals and the corresponding analogue signals all at the same time. (See, “Digital Designers Grapple with Analog Mixed Signal Designs”) Further, the industry now provides analogue IP blocks for tighter integration with the digital design. The ultimate example is when you now take a radio hard macro and place it in a digital design to give your IoT SoC connectivity.

Blyler: As mixed signal and RF elements are integrated into the SoC, low power will be a key constraint. What tools and techniques are available to help in the architectural and low-level design of low power devices? What new challenges do IoT designs present that are unique from other spaces?

Kasser: All of the diverse IoT applications and use-cases are inevitably challenged by meager energy budgets. As a result, low-power design is a prerequisite and proper energy management must be addressed at the system level. An IoT wireless sensor-node SoC, for example, must be capable of working with a range of operating modes that switch internal resources on and off as necessary to reduce energy consumption to sometimes even less than 1uA in idle modes. To get there, the chip must often be fitted with multiple power islands, various forms of body bias, retention modes, and other tricks. The tradeoffs these techniques require (chip size, complexity, active vs. leakage power, etc.) must be evaluated against various technology options that might be incompatible with integrating RF blocks efficiently.

New IoT chips must incorporate more processing and sophisticated RF communication capability (beaconing, wakeup modes, higher throughputs, etc.). As a result, they present unique challenges because they are asked to perform substantial more data crunching–at roughly the same small energy budget–that the simple signal conditioning performed by traditional sensor nodes.

Martin: Almost all nodes / motes are battery powered, perhaps even just through OTA (Over The Air) Energy harvesting.  This is adding additional pressure to the chip designers to modify existing analog blocks to become even more power efficient. Digital blocks tend to be easier since the bulk of the power is consumed during active clock cycles.  Once of the new challenged introduced in the IoT space is the low power RF signal strength problem associated with personal health monitoring devices which unless cell phones and their relatively high power RF sections need to receive and transmit data at very low power levels while being strapped directly to a 150 pound bag of water.

Parmenter: These techniques have been honed in portable applications and satellite — even automotive applications — for years. We will see these techniques used in other areas, like IoT, now that they are needed. The ability to monitor system performance and decide what to power up or down to conserve power has been done before. Digital control of power supplies will help with this as well. In applications such as energy harvesting and the monitoring of infrastructure – bridges and roads for example — we will need highly efficient energy harvesting devices. Super capacitors could store energy in a lossless fashion. Low power RF could send the data to “the cloud.” .

Figure: Good ULP for energy harvesting applications have an MCU with a low amount of active time. (Courtesy of Mouser)

Smith: Low power is an assumed requirement for almost all SoCs today. For mixed-signal, these techniques are mostly handled in the digital domain, but overall the digital and analog parts must still be designed and verified together to ensure functionality. Among many challenges that are perhaps unique to other spaces is that the devices may need to be placed in harsh conditions due to their need to be in close proximity to the host for sensing or location applications. As a result, reliability across wide ranges of temperature, power sources, electromagnetic interference levels and other environmental variables will need to be analyzed and verified to have appropriate tolerance levels. EDA tools are essential to ensuring these variations in environmental conditions do not cause the designs to fail in the field.

Soubeyrat: Diffusion of communicating objects on a large scale for IoT applications will require low cost sensors and processors combined with ultra-low power designs. One example is high grade motion tracking from low grade (hence low cost) sensors. The sensor data compensates for the long term drift of low cost gyroscopes by fusing their data with data from low grade accelerometer and magnetometer sensors. On the low-power side, power for the sensors could be managed by switching on the strict minimum of sensors and the power of the processing unit (MCU) by implementing an optimal duty cycle for the data fusion.

Soubra: Today’s existing SoC tool flows support all sorts of low power techniques at the gate and block level, i.e., power domains, state retention, and others. Gating of the clocks to peripherals via the creation of sophisticated clock trees is also very well understood and implemented by the tools. The part that is missing is enabling low power at the software level, to allow the system designer to incorporate them early into the design of the application. The software application must be designed to power down the peripheral when not in use, e.g., by calling on the clock gating function. The application must put the processor or analogue block to sleep when all events have been handled. A tight coupling between low power features in silicon and low power aware design in software yields the ultimate low power design of the system.

Blyler: The third element of an IoT strategy is security. How is the design of secure systems different in an IoT environment– from a MCU and system software-hardware context? What are the best tools and practices when designing long-term secure IoT systems?

Kasser: There is no fundamental difference to security in an IoT system than might be found in another kind of system. The approach requires identifying the assets, threats, and risks for the particular application/service/use case and then applying the right mix of state-of-the-art hardware and software security techniques/technologies to reach the target protection level and performance. One priority (and challenge) is to restrict the interfaces (attack surface) to the minimum possible, and apply state-of-the-art robust hardware and software design and fire-walling techniques to eliminate (best case) or at least minimize the impact of implementation-related vulnerabilities. This is particularly important for complex hubs and gateways. For nodes, conflicting security and low-power requirements require careful analysis and experience to achieve optimal system performance.

Martin: Security in the IoT space can be split into two areas, which in the general discussion of security will always overlap. Firstly, the IoT device space suggests a mind numbingly large number of nodes distributed around the house, car, personal space and everywhere.  This mandates that software updates be performed the RF channel much like cell phones are today.  However an extra layer of fault tolerance needs to be added to ensure that the software update is indeed legitimate and that there is a fall back mechanism to recover a bad image, either because it’s tainted or through a hardware fault in programming the new image.  Microcontroller manufacturers are addressing these concerns by adding hardware encryption blocks, dual bank flash and external devices that allow for very specific checks against contaminated code.  Best practices in this space are carries over from the fault tolerant industrial space which include SHA256 digest calculation and comparison on new software downloads.

Secondly, the content of the node data itself may or may not contain personal information, and it many cases the actual payload sizes are small to keep over all power consumption low.  It is however important that the RF channel be secured specifically when commands are sent to the nodes..  It’s perhaps not a large issue if your neighbor knows that your living lights are on but it’s a far different problem if your neighbor can control your living room lights.  Best practices in this space will include specific and pseudo (or completely ) random challenge response sequences to nodes before the actual commands are sent to ensure that the target node is indeed the correct recipient and that the IoT gateway is indeed the authorized control node.

Parmenter: I can’t think of a larger target for hackers and terrorists than accessing infrastructure to control, corrupt, steal or destroy data.  The value of the data and the system will far outweigh the hardware. Thus, it might be better to buy something that is proven in other high security applications rather than trying to write something on your own because it can save money. Further, requirements that exist in the military and aerospace market will apply to these systems for robustness and security. I believe the software and I/O are going to be under close scrutiny such that they cannot be compromised, yet allow simple access by authorized users.

Soubra: The main difference in security design for the IoT is the “I,” – the Internet. Before IoT, the system was either standalone or in a closed network. Within that context, security design is much simpler. But within the open network of the Internet, designers are now faced with unlimited security threats that increase every day in variety and style. Engineers must design the endpoint to protect itself against today’s attacks and supporting updates for tomorrow’s attacks. An even better approach would be to participate with the other nodes to interactively report suspicious activity on the network. Of course, all of this must be done in a very constraint code space and power budget!

Blyler: Thank you!

Tags: , , , , , , , , , , , , , ,

Leave a Reply