headache material is here, but give it a try...

http://chipdesignmag.com/sld/blog/2014/12/02/iot-cookbook-analog-and-digital-fusion-bus-recipe/

• Home
• Join the Community
• News
• Blogs
• Newsletter
• Chip Design
• Low-Power Engineering
• SemiMD

Headlines

• Blog Review - Monday, December 01 2014

Headlines

IoT Cookbook: Analog and Digital Fusion Bus Recipe

Experts from ARM, Mathworks, Cadence, Synopsys, Analog Devices, Atrenta, Hillcrest Labs and STMicroelectronics cook up ways to integrate analog with IoT buses.

By John Blyler, Editorial Director

Many embedded engineers approach the development of Internet-of-Things (IoT) devices like a cookbook. By following previous embedded recipes, they hope to create new and deliciously innovative applications. While the recipes may be similar, today’s IoT uses strong concentration of analog, sensors and wireless ingredients. How will these parts combine with the available high-end bus structures like ARM’s AMBA? To find out, “IoT Embedded Systems” talked with the head technical cooks including Paul Williamson, Senior Marketing Manager, ARM; Rob O’Reilly, Senior Member Technical Staff at Analog Devices; Mladen Nizic , Engineering Director, Mixed Signal Solution, Cadence; Ron Lowman, Strategic Marketing Manager for IoT, Synopsys; Corey Mathis, Industry Marketing Manager -  Communications, Electronics and Semiconductors, MathWorks; Daniel Chaitow, Marketing Manager, Hillcrest Labs; Bernard Murphy, CTO, Atrenta; and Sean Newton, Field Applications Engineering Manager, STMicroelectronics. What follows is a portion of their responses. — JB

Key points:

• System-level design is needed so that the bus interface can control the analog peripheral through a variety of modes and power-efficient scenarios.
• One industry challenge is to sort the various sensor data streams in sequence, in types, and include the ability to do sample or rate conversion.
• To ensure the correct sampling of analog sensor signals and the proper timing of all control and data signals, cycle accurate simulations must be performed.
• Control system and sensor subsystems are needed to help reduce digital bus cycles by tightly integrating the necessary components.
• Hardware design and software design have inherently different workflows, and as a result, use different design tools and methodologies.
• For low-power IoT sensors, the analog-digital converter (ADC) power supply must be designed to minimize noise. Attention must also be paid to the routing of analog signals between the sensors and the ADC.
• Beyond basic sensor interfacing, designer should consider digitally assisted analog (DAA) – or digital logic embedded in analog circuitry that functions as a digital signal processor.

Blyler: What challenges do designers face when integrating analog sensor and wireless IP with digital buses like ARM’s AMBA and others?

Williamson (ARM): Designers need to consider system-level performance when designing the interface between the processor core and the analog peripherals. For example a sensor peripheral might be running continuously, providing data to the CPU only when event thresholds are reached. Alternatively the analog sensor may be passing bursts of sampled data to the CPU for processing.  These different scenarios may require that the designer develop a digital interface that offers simple register control, or more advanced memory access. The design of the interface needs to enable control of the peripheral through a broad range of modes and in a manner that optimizes power efficiency at a system and application level.

O’Reilly (Analog Devices): One challenge is ultra-low power designs to enable management of the overall system power consumption. In IoT systems, typically there is one main SoC connected with multiple sensors running at different Output Data Rates (ODR) using asynchronous clocking. The application processor SoC collects the data from multiple sensors and completes the processing. To keep power consumption low, the SoC generally isn’t active all of the time. The SoC will collect data at certain intervals. To support the needs of sensor fusion it’s necessary that the sensor data includes time information. This highlights the second challenge, the ability to align a variety of different data types in a time sequence required for fusion processing. This raises the question “How can an entire industry adequately sort the various sensor data streams in sequence, in types, and include the ability to do sample or rate conversion.?”

Nizic (Cadence): Typically a sensor will generate a small (low voltage/current) analog signal which needs to be properly conditioned and amplified before converting it to digital signal sent over a bus to memory register for further processing by a DSP or a controller. Sometimes, to save area, multiple sensor signals are multiplexed (sampled) to reduce the number of A2D converters.

From the design methodology aspect, the biggest design challenge is verification. To ensure analog sensor signals are sampled correctly and all control and data signals are timed properly, cycle-accurate simulations must be performed. Since these systems now contain analog, in addition to digital and bus protocol verification, a mixed-signal simulation must cover both hardware and software. To effectively apply mixed-signal simulation, designers must model and abstract behavior of sensors, analog multiplexers, A2D converters and other analog components. On the physical implementation side, busses will require increased routing resources, which in turn mean more careful floor-planning and routing of bus and analog signals to keep chip area at minimum and avoid signal interference.

Lowman (Synopsys): For an IC designer, the digital bus provides a very easy way to snap together an IC by hanging interface controllers such as I2C, SPI, and UARTs to connect to sensors and wireless controllers.  It’s also an easy method to hang USB and Ethernet, as well as analog interfaces, memories and processing engines.  However, things are a bit more complicated on the system level. For example, the sensor in a control system helps some actuator know what to do and when to do it.  The challenge is that there is a delay in bus cycles from sensing to calculating a response to actually delivering a response that ultimately optimizes the control and efficiency of the system.  Examples include motor control, vision systems and power conversion applications. Ideally, you’d want a sensor and control subsystem that has optimized 9D Sensor Fusion application. This subsystem significantly reduces cycles spent traveling over a digital bus by essentially removing the bus and tightly integrating the necessary components needed to sense and process the algorithms. This technique will be critical to reducing power and increasing performance of IoT control systems and sensor applications in a deeply embedded world.

Mathis (Mathworks): It is no surprise that mathematical and signal processing algorithms of increasing complexity are driving many of the innovations in embedded IoT. This trend is partly enabled by the increasing capability of SoC hardware being deployed for the IoT. These SoCs provide embedded engineers greater flexibility regarding where the algorithms get implemented. The greater flexibility, however, leads to new questions in early stage design exploration. Where should the (analog and mixed) signal processing of that data occur? Should it occur in a hardware implementation, which is natively faster but more costly in on-chip resources? Or in software, where inherent latency issues may exist? One key challenge we see is that hardware design and software design have inherently different workflows, and as a result, use different design tools and methodologies. This means SoC architects need to be fluent in both C and HDL, and the hardware/software co-design environments needed for both. Another key challenge is that this integration further exacerbates the functional, gate- or circuit-level, and final sign-off verification problems that have dogged designers for decades. Interestingly, designers facing either or both of these key challenges could benefit significantly from top-down design and verification methodologies. (See last month’s discussion, “Is Hardware Really That Much Different From Software?”)

Chaitow (Hillcrest Labs): In most sensor-based applications, data is ultimately processed in the digital realm so an analog to digital conversion has to occur somewhere in the system before the processing occurs. MEMS sensors measure tiny variations in capacitance, and amplification of that signal is necessary to allow sufficient swing in the signal to ensure a reasonable resolution. Typically the analog to digital conversion is performed at the sensor to allow for reduction of error in the measurement. Errors are generally present because of the presence of noise in the system, but the design of the sensing element and amplifiers have attributes that contribute to error. For a given sensing system minimizing the noise is therefore paramount. The power supply of the ADC needs to be carefully designed to minimize noise and the routing of analog signals between the sensors and the ADC requires careful layout. If the ADC is part of an MCU, then the power regulation of the ADC and the isolation of the analog front end from the digital side of the system is vital to ensure an effective sampling system.

As always with design there are many tradeoffs. A given analog MEMS supplier may be able to provide a superior measurement system to a MEMS supplier that provides a digital output. By accepting the additional complexity of the mixed-signal system and combining the analog sensor with a capable ADC, an improved measurement system can be built. In addition if the application requires multiple sensors, using a single external multiple channel ADC with analog sensors can yield a less expensive system, which will be increasingly important as the IoT revolution continues.

Murphy (Atrenta): Aside from the software needs, there are design and integration considerations. On the design side, there is nothing very odd. The sensor needs to be presented to an AMBA fabric as a slave of some variety (eg APB or AHB), which means it needs all the digital logic to act as a well-behaved slave (see Figure). It should recognize it is not guaranteed to be serviced on demand and therefore should support internal buffering (streaming buffer if an output device for audio, video or other real-time signal). Sensors can be power-hungry so they should support power down that can be signaled by the bus (as requested by software).

The implementation side is definitely more interesting. All of that logic is generally bundled with the analog circuitry into one AMS block and it is usually difficult to pin down a floor-plan outline on such a block until quite close to final layout. This makes full-chip floor planning more challenging because you are connecting to an AMBA switch fabric, which likes to connect to well-constrained interfaces because the switch matrix itself doesn’t constrain layout well on its own. This may lead to a little more iteration of the floor plan than you otherwise might expect

Beyond basic sensor interfacing, you need to consider digitally assisted analog (DAA). This is when you have digital logic embedded in analog circuitry, functioning as a digital signal processor to perform effectively an analog function but perhaps more flexibly and certainly with more programmability that analog circuitry. Typical applications are for beamforming in radio transmission and for super-accurate ADCs.

Figure: The AMBA Bus SOC Platform is a configurable with several peripherals and system functions, e.g., AHB Bus(es), APB Bus(es), arbiters, decoders. Popular peripherals include RAM controllers, Ethernet, PCI, USB, 1394a, UARTs, PWMs, PIOs. (Courtesy of ARM Community - http://community.arm.com/docs/DOC-3752)

Newton (STMicroelectronics): Integration of devices such as analog sensors and wireless IP (radios) is widespread today via the use of standard digital bus interfaces such as I2C and SPI. Integration of analog IP with a bus – such as ARM’s AMBA – becomes a matter of connecting the relevant buses to the digital registers contained within the IP. This is exactly what happens when you use I2C or SPI to communicate to standalone sensors or wireless radio, with the low-speed bus interfaces giving external access to the internal registers of the analog IP. The challenges for integration to devices with higher-end busses isn’t so much on the bus interface, as it is in defining and qualifying the resulting SoC. In particular, packaging characteristics, the number of GPIO’s available, the size of package, the type of processing device used (MPU or MCU), internal memory capability such as flash or internal SRAM, and of course the power capabilities of the device in question: does it need very low standby power? Wake capability?  Most of these questions are driven by market requirements and capabilities and must be weighed against the cost and complexity of the integration effort.

The challenges for integration to devices with higher-end busses isn’t so much on the bus interface, as it is in defining packaging characteristics, available GPIOs, type of processing device, memory such as flash or internal SRAM, and power capabilities.