Differentiating IoT hardware
05 October 2016
Today’s connected products for smart cars, smart home, smart cities, and other areas of the IoT, must perform with an ‘always-on’ state of readiness.
(Click here to view article in digital issue)
This presents clear challenges to designers of these devices in meeting power and performance targets.
We see IoT products falling into three distinct classes depending on the type of data they handle: Machine to Machine (M2M), Audio, and Audio/Video. It is beneficial to categorise the requirements for IoT devices in this way in order to better understand the power consumption, power management and processing performance requirements.
Imagination Technologies has developed solutions to help engineers with the key processing blocks needed to create SoCs that will thrive in IoT environments. These scalable IP solutions have been designed to support every level of IoT device and solve common issues including managing wireless communications bandwidth, providing sufficient processing resources to enable rich graphical UIs, video and images, and controlling power consumption and efficiency, as well as offering maximum levels of network and data security.
Tackling power management
For edge products on the network and specifically sensor devices, power management is clearly of concern for most devices. Many of these devices are intended to be operated by a small battery for extended periods of time as measured in years. In a battery powered device, it is imperative to optimise both dynamic and static power consumption. Power optimisation is addressed in three different ways:
• Power management control
• IP implemented for low power
• Power aware software
Power management control refers to the policy setting mechanism for the various functions inside the SoC. The power management policy defines the power states that specific functions of the SoC enter depending on the external inputs as well as software. The power states as shown in Figure 2 are set in part by voltage scaling and frequency scaling of the specific functional blocks within the SoC.
IP blocks for IoT SoCs should be designed to include power control wrappers for power and frequency scaling as shown in Figure 2. Control of the power states of the functional blocks of the SoC is accomplished through the chip fabric.
IP providers like Imagination support power state control of the IP block. To implement IP for low power operation, the system designer must first identify the power management objectives for the SoC. In the case of an IoT SoC, where the device is turned off for significantly longer time periods than it is turned on, leakage power will dominate the power consumption. In this scenario, techniques to minimise leakage are required including:
• Use of power gating. By gating power to a block of logic, the active area of that logic block that is directly connected to the supply voltage is reduced. Leakage power is directly related to the area of the chip that is connected to power and can form a reverse biased diode to ground.
• Use of reduced leakage memories. When memory blocks are not accessed, the SoC designer will want memories that can retain their information while remaining in a low power state. This is typically accomplished in memories in which the memory array power supply voltage can be independently reduced to a minimum voltage that will guarantee memory retention when the memory is not accessed.
• Use of high Vt logic blocks to reduce leakage. Provided that the logic blank can meet its performance target, high Vt standard cells will exhibit lower leakage than standard Vt standard cells.
If the SoC is turned on for the majority of time, as in IoT devices such as sensor hubs, dynamic power will dominate. To reduce the dynamic power, voltage and frequency scaling should be implemented as a part of the power management function. Reducing the supply voltage for digital logic has a significant impact as PDYNAMIC a CV2f where:
C = loading capacitance on output of CMOS output;
V = supply voltage;
F = switching frequency.
A useful scheme to reduce power for a CPU is to close timing based on reduced values of supply voltage. This is commonly referred to as voltage scaling. For example, operating a CPU through voltage scaling at 0.95 V as opposed to the minimum voltage of 1.08 V results in a power reduction of 23%.
Bluetooth Smart (also known as Bluetooth Low Energy or BLE) is positioned for very low-power wireless communications, but the power reduction comes at the cost of reduced range point-to-point communications, and low data rates. For applications requiring higher data rates (see Figure 3), Wi-Fi would be a suitable solution. Bluetooth and 802.11n can handle data requirements with the exception of requirements for streaming HD video.
Imagination has developed a low-power Wi-Fi and Bluetooth offering with baseband and RF called the ‘Whisper’ RPU to tackle this. The low-power Wi-Fi baseband is integrated with AFE (analogue front end) and RF transceiver to form a complete solution for integrating 802.11n wireless connectivity. Whisper achieves low power by exploiting the low power aspects within the 802.11 specification, process node, and IoT use cases. The integration of the MIPS processor with the Whisper RPU provides power advantages at the Wi-Fi system level by optimising wake up and sleep cycles.
The IEEE 802.11 a/b/g/n standards also provide a number of features to drive lower power in Wi-Fi SoCs including IEEE Power Save Mode. In this mode, radio activity is suspended after a period of inactivity to save power. The SoC wakes up periodically to check if the access point has data queued for it to ensure no valid packets are lost. In Wireless Multimedia (WMM) Power Save mode, the client can request for queued traffic any time, rather than waiting for the next beacon frame.
The Ensigma Whisper RPU architecture uses a combination of techniques discussed above, as well as several PHY, MAC and RF optimisations to attain the lowest power consumption possible.
Whisper RPUs also go further with specific optimisations for IoT. In general, IoT SoCs are asleep most of the time; they wake up only once in a while to determine if a packet needs to be decoded or transmitted. Given these considerations, Whisper RPUs are designed to optimise the sleep mode power consumption.
Figure 5 showcases the multiple power states in Whisper and compares the same with a direct competitor. As is evident from the figure, the receive power consumption is roughly half that of the competition. The various power optimisations ensure that Whisper-powered SoCs can wake up quickly, consuming the lowest power in the process. These power-sipping characteristics make Whisper RPUs highly suitable for IoT applications running on coin cell batteries.
CPU processing performance for IoT
Architectural considerations for the CPU performance in an IoT SoC will depend on the scope of what the CPU needs to do, as well as hardware security provisions contained within the hardware of the CPU.
In Figure 5, a MIPS M5150 CPU which includes hardware virtualisation is a good fit, providing the performance needed to run sensor interpretation code, a Wi-Fi stack, and internet communications at a clock speed of about 100MHz. Most IoT SoCs will be designed with additional CPU capability to support extra features that can be added via software upgrades. As a result, the performance of the CPU will need to be scalable, and may also include special pipeline stages to include special purpose processing that is deemed necessary to be done locally.
The CPU chosen for a specific IoT application must not only support the security features, but must also be scalable in performance by implementation to support higher clock frequencies. For certain applications, it is also beneficial for the CPU to support hardware multi-threading as found in CPUs such as MIPS I-class processors).
Typical wireless IoT SoCs will be designed to meet specific standards. The standards deployed will depend on the security requirements needed, the type of network topology to be supported (e.g. IP, mesh), and the data rates to be supported. The diagram below provides a classification of IoT network requirements based on sustained data rates. Since Wi-Fi is pervasively deployed today, most IoT applications will support Wi-Fi. Also, for LED lighting and applications that may span large geographic areas, ZigBee networks are used and may be present in IoT systems alongside Wi-Fi.
Aside from HD video streaming applications such as those used in home entertainment or security video monitoring, 802.11n and 802.11n 1×1 provide sufficient bandwidth; the 802.11 networks will use dual band 2.4GHz and 5.5GHz frequencies. For lower power and lower cost implementations, 802.11n can be supported by a single band 2.4 GHz radio. The lower frequency bands are more desirable since the RF transmission will provide greater range for a given power level output.
Wireless communications will become integrated into the main SoC not only to reduce cost, but also to reduce power consumption and improve system performance. The inclusion of the wireless connectivity and its associated software stacks, integrated capabilities to do some local analytics, and security will increase the demand for computational power within the SoC.
A key requirement for IoT applications is security. IoT opens up networks to a variety of threats as more and more products are connected to a network and eventually to the cloud. Figure 6 shows an example of an IoT product for home automation connected to a home network with possible threats to the network security.
At the edge of the network, as multiple IoT products are added, the potential threats are greatly increased. IoT devices must be capable of providing a robustly secure environment. Strategies to improve security include:
• Secure boot
• Secure code update
• Key protection
• Tamper resistance
• Access control of secure resources
• DPA and side channel attack resistance
• Secure DMA (direct memory access) with data encryption for critical functions
• Session authentication
The security solution needed for an IoT application is dependent on what needs to be protected. For example, in the case of a connected lightbulb, the main security threat would be to the network that the lightbulb is attached. In the case of a wearable medical device, the data as well as the network will need to be protected. Imagination’s Omnishield security technology provides the foundation for scalable multi-domain security which is enabled by Imagination IP.
As IoT products become pervasive in networked systems, the SoC providers for these systems will need to differentiate their products based on security, power management, scalable computational performance and compliance to industry driven standards. Imagination’s CPU, GPU, communications, video and imaging IP cores are designed to meet the most aggressive requirements and opportunities for product differentiation in IoT applications.
Contact Details and Archive...