Physical AI systems are moving from development platforms into production deployments. Humanoid robots, autonomous mobile robots, manipulators, smart infrastructure, and industrial systems all require sufficient AI performance, but peak compute alone does not determine system capability. Production designs also need to guarantee deterministic sensor latency, manageable power consumption, and a hardware architecture that can scale across multiple product configurations and can be cost optimized. 

NVIDIA is addressing the compute side of that requirement by expanding the NVIDIA Jetson Thor family with the Jetson T3000 and T2000. Both modules use the same NVIDIA Blackwell architecture and Jetson software environment as the flagship T5000, but target smaller and lower-power systems. NVIDIA Holoscan Sensor Bridge addresses the other side of the architecture by transporting sensor data over Ethernet with PTP synchronization and a low-latency data path to GPU memory. Together, these technologies provide a practical way to scale physical AI systems without redesigning the complete compute and sensor stack for each performance tier. 

Tauro’s implementation: one sensor architecture across Thor platforms 

Tauro Technologies provides the production hardware layer for this architecture. Developed in close collaboration with NVIDIA and Lattice Semiconductor, our DA322 Holoscan MIPI Adapter implements the Holoscan Sensor Bridge technology in a compact production module. DA322 includes four MIPI CSI-2 inputs, 10GbE over SFP+, a Lattice CertusPro-NX FPGA with RDMA and GPUDirect support, IEEE 1588 PTP synchronization, and a compact footprint of 75mm x 45mm. 

The DA322 is part of a broader Tauro platform and sensor ecosystem: 

• TT317 IGX Thor platform – an enterprise-class edge AI platform for industrial, medical, and high-performance physical AI deployments. 

• TT310, TT314, TT315, and TT316 – rugged and industrial Jetson Thor platforms for outdoor robotics, autonomous systems, transportation, and high-throughput edge AI applications. 

• DA326 Holoscan GMSL Adapter – connects GMSL2 and GMSL3 cameras to NVIDIA platforms using Holoscan Sensor Bridge. 

• DA328 Holoscan 3G-SDI Adapter – connects SDI video sources to NVIDIA platforms using Holoscan Sensor Bridge. 

Tauro is also preparing platform support for the NVIDIA Jetson T3000 and T2000. Initial application and sensor-pipeline evaluation can begin with T3000 emulation on the NVIDIA Jetson AGX Thor Developer Kit. 

Each product shares the same FPGA baseline and software stack, making adding a new sensor type or stepping between Thor tiers with no design change which dramatically reduces NRE and shortens the path to system qualification. The same architecture serves AGVs and humanoid robotics, industrial automation, transportation, automotive, and medical systems: production-qualified, OEM-ready, and USA-manufactured. 

A scalable Jetson Thor architecture 

The main architectural advantage is consistency across the NVIDIA Jetson Thor family. T2000, T3000, T4000 and T5000 share the same NVIDIA Blackwell GPU architecture and the NVIDIA JetPack software stack. This allows engineering teams to evaluate workloads on one Thor platform and move between module tiers with substantially less software rework than a change to a different compute architecture. 

NVIDIA Jetson T3000 targets mainstream robotics and Edge AI that do not require the full memory capacity or power envelope of T5000. Preliminary specifications include 865 FP4 TFLOPS, an 8-core Arm Neoverse CPU operating at 2.6 GHz, 32 GB of LPDDR5X memory, x1 25GbE Ethernet Interface, and a 70 W module power target. NVIDIA positions T3000 at similar inference performance of the T5000 for selected LLM, VLM, VLA, and world-model workloads, while using roughly half the module power and a smaller form factor. 

NVIDIA Jetson T2000 provides a lower-power entry point to the Thor architecture. It combines 400 FP4 TFLOPS, a 6-core Arm Neoverse CPU at 2.6 GHz, 16 GB of LPDDR5X memory, and a 40W power target. This configuration is intended for applications such as autonomous mobile robots, manipulators, visual AI agents, and compact edge systems that need Blackwell-class inference capability within a constrained thermal and power budget. 

Because the modules use the same software foundation, applications can be developed with JetPack and NVIDIA’s physical AI software ecosystem, including NVIDIA’s open models such as Isaac GR00T, Nemotron, and Cosmos. The common architecture does not eliminate platform validation, thermal design, or I/O integration, but it reduces the amount of application-level rework required when moving between Thor compute tiers. NVIDIA is also bringing Cosmos 3 Edge , first full omnimodel for physical AI to Jetson Thor for real-time multi-view and livestream understanding and custom robot policy deployment at the edge. 

Jetson Thor modules at a glance 

Preliminary specifications. Subject to change. 

Scalable sensor platform as key part of the architecture 

A physical AI system is limited by how reliably it can acquire and process sensor data. Cameras, LiDAR, Radar, IMUs, encoders, and other sensors must reach the compute platform with predictable latency and a common time base. As sensor count and bandwidth increase, the sensor ingest architecture can become a system bottleneck even when the GPU has sufficient inference performance. 

NVIDIA Holoscan Sensor Bridge provides an alternative architecture. An FPGA-based sensor endpoint receives native sensor interfaces such as MIPI CSI-2, I2C, SPI, GPIO, and related protocols, packetizes the data for Ethernet transport, and delivers it to the host through the Holoscan software stack. With RDMA and GPUDirect, Holoscan Sensor Bridge allows delivering sensor payloads directly into GPU-accessible memory. NVIDIA-published benchmarks report signal-processing latency below 1 ms, approximately 1% CPU utilization for accelerated ingest, up to 10x lower latency than selected traditional pipelines, and significantly lower driver-development effort. Actual results depend on the sensor, network configuration, host platform, and processing pipeline. 

Why HSB matters when right-sizing compute 

The addition of T3000 and T2000 provides the ability to select a compute module based on the actual inference, memory, power, and thermal requirements of the application. That approach is effective only when sensor ingest does not consume a disproportionate share of the CPU and memory bandwidth. By moving sensor interface handling and packetization to the FPGA endpoint, and by using an accelerated host path, HSB reduces the amount of compute spent on sensor transport and buffer management. The recovered resources remain available for perception, planning, reasoning, and control workloads. 

The same sensor architecture can also be reused across NVIDIA AGX, IGX Orin, Jetson Thor, IGX Thor, and x86 systems with NVIDIA GPUs. The host integration still requires platform-specific validation, but the sensor endpoint, Ethernet transport, synchronization method, and application-facing data path remain consistent. On NVIDIA Jetson Thor and IGX Thor, the accelerated networking path is integrated into the platform. On Jetson AGX Orin, GPUDirect RDMA can be enabled through a compatible external NVIDIA ConnectX NIC. This common architecture makes it easier to prototype on one platform and deploy on another without redesigning the sensor interface. 

Do more with less: memory optimization and agent skills 

Hardware selection and software optimization must be considered together. Jetson systems use unified memory, so the operating system, inference runtime, models, pre-processing, post-processing, and sensor buffers all compete for the same memory pool. NVIDIA’s optimization guidance includes headless operating-system configurations, efficient runtimes such as llama.cpp and TensorRT Edge-LLM, and lower-precision formats including NVFP4 and INT4. Depending on the starting configuration, these changes can recover 10 GB or more of memory and allow larger or additional models to run on the same module. 

The Reachy Mini Jetson Assistant is one example of this approach. A local voice-and-vision agent combines a vision-language model, speech-to-text, text-to-speech, face tracking, and robot control on an 8 GB Jetson platform. NVIDIA’s Jetson agent skills are intended to inspect memory use, identify bottlenecks, and validate optimizations on the target hardware. Combined with low-overhead sensor ingest, these methods allow engineering teams to select a smaller Thor module without reducing application capability. 

Start designing soon – emulation available by end of the month 

The NVIDIA Jetson AGX Thor Developer Kit can emulate the T3000 and T2000 configurations while running the JetPack software stack, supported models, and Jetson agent skills. T3000 emulation is planned to be available by end of this onth , with T2000 emulation following later. This allows teams to characterize model performance, memory use, sensor bandwidth, and application latency before production modules are available in Q1 2027. 

The takeaway 

NVIDIA Jetson T3000 and T2000 extend the Blackwell-based Thor architecture into lower-power and lower-memory system configurations. NVIDIA Holoscan Sensor Bridge provides a common sensor-ingest architecture that can scale with those compute tiers by reducing host overhead, supporting PTP synchronization, and enabling accelerated data delivery to GPU memory. 

For engineering teams, the practical benefit is a more modular system architecture: select the Thor module based on the workload, then reuse the same sensor endpoint and transport architecture across multiple deployment platforms. Tauro provides the rugged compute platforms and production-ready HSB adapters required to implement that architecture. 

Design the Edge. Deliver the Future. 

The post Tauro Technologies Provides Platform Support for New NVIDIA Jetson T3000 and T2000 Modules  appeared first on Tauro Technologies.

In embedded systems, where software and hardware are tightly coupled, replacing obsolete hardware is rarely trivial. Whether due to component going EOL, performance shortfalls, or evolving application needs, the cost of re-engineering and re-qualifying systems from scratch can be substantial. Larger systems have years of design and validation testing prior to their release, and changes to mission-critical embedded systems are expensive and risky. 

That’s where software-compatible drop-in replacements come in – offering a smart, strategic solution that preserves system functionality, minimizes downtime, and avoids expensive revalidation cycles.

What Is a Software-Compatible Drop-In Replacement?

A software-compatible drop-in replacement hardware is engineered to mimic the form, fit, and function of the original device, while maintaining backward compatibility at the software and hardware interface level. These replacements integrate seamlessly into the existing systems without requiring changes to the interfacing hardware, firmware, or application software.

Achieving software-compatible hardware replacements requires a rigorous engineering process that ensures:

• Electrical and pin-level compatibility
• API and driver-level consistency
• Form factor and mechanical interface integrity
• System-level behavior across operating conditions

The compatibility ensures the replacement hardware behaves predictably within the larger system – allowing customers to extend the life of their platforms without rewriting the software stack or re-qualifying the hardware.

TauroTech’s Methodology for Developing Drop-In Replacements

In creating a drop-in replacement for an embedded system, an approach involving various engineering disciplines and extensive testing is imperative. Engineers must thoroughly understand the existing system to be able to design a matching replacement, build a prototype, and ensure it meets all requirements before moving into production.

Pre-Design Evaluation: Ensuring Technical and Functional Alignment

Evaluation is crucial for assessing the feasibility and the characteristics of new components, modules, or interfaces. This phase includes rigorous black-box functional system analysis and careful component selection.

Black-Box Functional System Analysis: This methodical approach is essential for understanding and evaluating system’s behavior and functionality without being concerned with its internal structure. Key points include defining functional requirements and operational characteristics through datasheets and documentation, mapping inputs to outputs, and examining system behavior under various usage models and corner-case scenarios.

Critical Component Selection: In this phase, it is imperative to thoroughly research available components in the market and select solutions that meet or exceed the specifications and performance of the original components. Critical components such as microprocessors, sensors, and power supplies must unequivocally meet performance and reliability criteria of the system being replaced.

Design Phase: Building a Compatible Replacement

The design phase focuses on creating a fully integrated drop-in replacement by aligning electrical, digital, mechanical, and software elements. The goal is to ensure the design fits, functions, and communicates just like the original – without requiring changes to the broader system.

Electrical Design: Engineers develop schematics and board layout that replicate the original hardware’s power, signaling, and interface requirements. Emphasis is placed on ensuring pin compatibility, stable power delivery, and reliable operation within the system.

FPGA and Digital Design: When programmable logic is needed, FPGAs or SoCs are selected based on performance, compatibility, and long-term support. Custom logic is developed to match system behavior and ensure seamless integration.

Mechanical Design: Mechanical aspects such as dimensions, mounting, and connector placement are carefully matched to the original. The design also accounts for thermal performance and reliability in the intended environment.

Software Development: Low-level software – including drivers and firmware – is developed or adapted to ensure the new hardware integrates seamlessly into the existing software stacks. This ensures system behavior remains consistent and reliable.

API and Interface Compatibility: User-facing APIs and communication protocols are preserved to avoid changes to application code. Compatibility at this level is key to minimizing integration time and risk.

System-Level Test & Form/Fit/Function Compatibility Validation 

Thorough validation at the system level is essential to ensure that drop-in replacements not only function independently but also operate reliably as part of the larger system.

System-level testing evaluates the replacement component in the context of the complete system, verifying correct behavior under real-world operating conditions. This process includes checking interoperability with other subsystems, confirming that timing, interfaces, and performance targets are met, and identifying any integration issues that might not be apparent during isolated component testing. For drop-in replacements, this step is critical to ensuring that legacy software and hardware continue to function as intended with the new module.

Form, fit, and function validation ensures that the replacement meets physical, mechanical, and functional expectations without requiring system modifications.

• Form confirms that the dimensions, shape, and footprint match the original.
• Fit ensures proper alignment with enclosures, connectors, and mounting points.
• Function verifies that the component performs its intended role within the system – electrically, thermally, and functionally – without impacting other components.

Design Validation and Acceptance Documentation

Design Validation Testing (DVT) and Acceptance Test Procedures (ATP) play complementary roles in the qualification and production of drop-in replacements, ensuring both the integrity of the design and the consistency of manufactured units.

DVT is performed during development to verify that the design meets all functional, electrical, mechanical, and environmental requirements. This includes defining the test environment, identifying key features to be validated, applying structured test methodologies, and confirming the design’s ability to meet performance specifications.

Acceptance Test Procedures (ATP) are implemented at the production stage to validate that each unit manufactured meets the defined quality and performance criteria. ATPs are derived from customer-approved specifications and simulate key aspects of real-world operation to confirm that the assembled product conforms to expectations. These procedures serve as a final gate before shipment and may include automated test sequences, functional checks, and go/no-go criteria for each production lot.

Lifecycle Engineering & Long-Term Support

Longevity of Supply: Longevity of supply is crucial for ensuring that products remain viable and supported in the market for their intended lifespan. TauroTech offers strategies to manage component availability, including notifying customers about product end-of-life (EOL) and suggesting options such as purchasing customer-bonded stock or updating the product in lock-step with customer evaluation and approval process. Bonded stock involves reserving inventory for a specific customer, ensuring their long-term supply needs are met, while revision-locked production ensures no software and design changes are made without the customer’s formal sign-off. This customer-inclusive approach aims to minimize supply chain disruptions and ensure that product changes correspond to their needs and standards before being applied.

Longevity of Repair: Longevity of repair in embedded systems refers to the expected duration of the repair capability, which is crucial in industries where products that are no longer produced may require maintenance or replacements. Factors such as spare part availability, manufacturer support, design features, and industry regulations need to be considered for decision-making. Tauro Technologies’ repair strategy is based on mutual agreements with customers, ensuring repair services and support for a specified period. To fulfill this commitment, Tauro Technologies holds customer-bonded stock, allowing for necessary repairs even when parts become unavailable. At the end of the contract, all stock is returned to the customer.

Roadmap Planning: Roadmap planning involves creating a structured plan to meet changing customer needs. Tauro Technologies ensures a smooth transition to new products, involving customers in decision-making processes and ensuring alignment with customer expectations. Tauro Technologies also supports the re-qualification process and offer guidance, focusing on integrating the replacement component by engaging customers and prioritizing their needs. 

Why Tauro Technologies?

With over 15 years of embedded system expertise and full in-house capabilities, Tauro Technologies is uniquely positioned to deliver high-reliability, software-compatible hardware replacements that are thoroughly tested to ensure reliability across all operational scenarios – including edge cases and extreme conditions.

We support customers in defense, medical, transportation, and robotics sectors with:

• Hardware replacements designed for Form/Fit/Function and Software compatibility
• Integrated electrical, mechanical, and firmware engineering under one roof
• Manufacturing through AS9100/ISO9001-certified partners
• Rapid prototyping and system-level validation for faster deployment

Looking to replace aging hardware with a fully compatible solution—backed by better supply chain control, and long-term support? Let’s talk.

📩 Reach out at info@taurotech.com or visit www.taurotech.com to start the conversation.

The post Software-Compatible Hardware Drop-In Replacements appeared first on Tauro Technologies.

An embedded system refers to a specialized computer system designed to perform dedicated functions within a larger mechanical or electrical system. It typically consists of a combination of hardware and software components tailored to perform specific tasks or functions. Embedded systems play a crucial role in mobile robotics, UAV construction and edge AI. Such systems are characterized by their real-time operation, reliability and efficiency in executing predetermined functions, often with limited resources such as processing power, memory and energy. In remote areas, for example, everything is run off batteries or generators. Consequently, many embedded systems are engineered to incorporate various techniques to extend battery life. The others simply need to consume less energy based on other factors. As a result, there’s an increasing demand for designs that minimize energy usage while maintaining high performance. In this article we are going to elaborate on the strategies for achieving low-power designs and highlight their significance in embedded systems.

The Need for Low-Power Design

Low-power design involves strategies and approaches aimed at decreasing the energy usage of electronic devices and refers especially to the underlying embedded systems upon which such devices operate. Examples of such devices are battery-powered devices, processors, IoT wireless sensor networks and many more. Through the application of low-power design methods, engineers can create high-quality and reliable equipment which consume considerably less energy without any indication of performance degradation. The need for low-power devices arises from several factors:

• Power sources are often limited and the disruption in the energy supply can result in adversities. This is particularly true for battery-powered devices in military situations where power outages can cost lives. That’s why defense is always looking for lower power consumption in airborne and ground vehicle applications.
• Portability of everyday devices (notebooks, smartphones, etc.) which will have prolonged battery life is one of the concerns of device manufacturing companies. In today’s world, it is a common tendency for customers to have a preference for devices with extended battery life.
• Low-power design will certainly have a huge positive impact on the environment as a large amount of electricity is wasted through devices connected to the grid. The decrease in the electricity consumption of such devices will result in less costs and will cause less damage to the environment.
• In embedded systems, high power consumption can result in a significant amount of heat generation damaging the system components. The reduction of generated heat is one of the concerns for military equipment production. As a fact, the overall decrease in power consumption will considerably reduce the generated heat. Consequently, initially employing low-power design techniques will protect the system from unexpected side effects due to thermal issues.
• Less heat generation can lead to improved performance and reliability of the embedded system. Overheating can cause performance degradation or even hardware failures, so by keeping temperatures within acceptable limits, low-power designs contribute to overall reliability and durability of the system.

Key Principles of Low-Power Design

To grasp the fundamental principles of low-power design, it’s imperative to dive into power consumption basics, sleep modes, clock gating techniques and voltage scaling strategies. This exploration will shed light on how each aspect contributes to the creation of energy-efficient embedded systems.

Power Consumption Basics

Power consumption indicates how much electrical energy a device or a system uses to perform its functions or operations. There are two primary sources of power consumption in electronic devices – static and dynamic. Devices consume static power when idle and dynamic power during active use. Reducing both static and dynamic power consumption is essential for creating low-power designs achieved through the means of efficient components and optimized circuits. Understanding consumption allows informed decisions on resource allocation and environmental impact mitigation. Embracing energy-efficient practices drives towards sustainability while ensuring reliable access to necessities.

Power Management and Sleep Modes

Implementing sleep modes and power states can significantly reduce power consumption in embedded systems. Sleep modes enable devices to enter low-power states when not performing tasks, therefore conserving energy. Power states define consumption levels based on system activity and performance needs. Selecting appropriate modes ensures optimal power usage and performance while maintaining efficiency.

All sleep modes are accessible from active mode, where the CPU executes application code. Upon entering sleep mode, program execution halts, and the device relies on interrupts or a reset for waking up. The application code determines the timing and choice of sleep mode. Enabled interrupts from peripherals and reset sources can return the CPU from sleep to active mode. Furthermore, power reduction registers offer means to halt individual peripheral clocks via software control. This action freezes the peripheral’s current state, eliminating its power consumption. Consequently, power usage is minimized in both active mode and idle sleep modes, facilitating more nuanced power management than sleep modes alone.

Here are several examples of low-power modes:

Sleep Mode: In this mode, the device reduces its power consumption by powering down non-essential components while retaining data in memory. The CPU typically enters a low-power state, halting its operation until an external event, such as a button press or an interrupt, wakes it up.

Deep Sleep Mode: This mode is an even lower power state compared to sleep mode. In deep sleep, the device shuts down most of its non-essential functions, including reducing power to the CPU and peripherals. This mode is commonly used in battery-powered devices to prolong battery life during extended periods of inactivity.

Standby Mode: This mode is similar to sleep mode but may involve a slightly higher level of power consumption. In this mode, the device reduces power to most components, but some essential functions remain active to enable quick recovery. It’s commonly used in devices like TVs and remote controls, where rapid responsiveness is necessary.

Clock Gating for Dynamic Power Reduction

Clock gating is a technique aimed at reducing dynamic power consumption by selectively switching off unnecessary clock signals to registers using control signals, all while ensuring functional correctness. By turning off the clock to idle parts of a device, it conserves power, directing it only to active components and minimizing waste. Implementing clock gating in embedded systems can substantially reduce power usage, particularly in devices with numerous components or intricate functionalities.

Typically, the assignment to a register might be conditional, as depicted above. When EN is 0, the clocks to the registers can be stopped otherwise, the registers will switch states on each clock cycle, which dissipates power.

Voltage Scaling Strategies

Voltage scaling strategies in low-power design involve adjusting the core supply voltage to align with the system’s performance needs. Decreasing voltage decreases power consumption, but it can impact performance, necessitating a careful balance between the two. Techniques like adaptive voltage scaling and dynamic voltage scaling are commonly used in embedded systems to find this balance, often coupled with frequency scaling to maintain acceptable performance levels while reducing power consumption. Dynamic Voltage and Frequency Scaling (DVFS) is a power management technique that adjusts the voltage and frequency of the device’s CPU dynamically based on workload demands. During periods of low activity, the CPU voltage and frequency are decreased to save power, while they are increased during high-demand tasks to maintain performance. These strategies are particularly crucial in portable devices where battery life is a primary concern.

Design Techniques for Low-Power Embedded Systems

When thinking about the low-power embedded systems, there is no single rule that applies to every type of requirement. Rather it is a combination of a system design, circuit design and firmware design all combined and working together to deliver the best performance per watt. Embedded engineers construct embedded systems using various low-power techniques, allowing for adaptable control over device’s energy usage based on its activities and operating patterns. 

Hardware Techniques for Low-Power Design

In the realm of low-power embedded system design, the selection of hardware components plays a pivotal role. Optimal choices can significantly influence the system’s overall power consumption. This section will delve into various hardware techniques, such as component selection for low-power embedded systems, employing energy-efficient microcontrollers and processors, and integrating sensors designed for minimal energy consumption.

Energy-efficient component selection: Picking the right components is crucial for any electronic system, affecting design, layout, and power usage. When it comes to low-power designs, choosing components wisely is even more critical. To reduce power consumption in embedded systems, we need to focus on factors like operating voltage, idle/standby current, and overall efficiency of the components. Opting for parts with lower consumption can significantly cut down on energy usage in the system.

Energy-efficient microcontroller and processor selection: Embedded systems rely heavily on microcontrollers and processors, and their power efficiency is crucial in determining overall power usage. When choosing a microcontroller or a processor, prioritize components with low operating voltages, effective sleep modes, and power-saving capabilities like clock gating and voltage scaling. Incorporating these features ensures decreased power consumption without compromising performance, making them ideal choices for energy-conscious designs.

One example of a low-power AI accelerator is Hailo-15 that can process multiple video streams in real time on a single device with robust onboard network connectivity. It offers very high AI performance of 26 TOPS and very low power consumption of 2.5W which makes it perfect for AI computing and for mission-critical applications with power consumption reduced by approximately 70% compared to GPU based solutions. Another example is Intel’s hybrid CPU architecture, which combines “P cores” for high-intensity computational tasks and “E cores” for handling less-intensive tasks while maximizing energy-efficiency, addressing the requirements of modern computing.

Energy-efficient process node selection: When talking about semiconductor ICs, selecting newer devices with 5nm technology node vs 10nm reduces power by 40%, 3nm improves 45% over 5nm, 14nm reduces power by 50% over 28nm etc. Power efficiency can be dramatically improved when using IC built on top of latest technology node.

Energy-efficient FPGA design: Field-Programmable Gate Array (FPGA) devices offer the advantage of flexibility and customization in hardware design. In certain applications, this flexibility can lead to power reduction by combining multiple functions into a single FPGA device rather than using discrete components.

Energy-efficient sensor selection: Sensors play a crucial role in embedded systems, gathering data from the surroundings or user interactions. Opting for sensors with minimal power demands that can transition into low-power modes when inactive is a key. Furthermore, explore sensors equipped with built-in power management functionalities like automatic sleep modes and adjustable sample rates to enhance energy efficiency even further. By selecting sensors with these capabilities, overall power consumption in the system can be significantly reduced, ensuring efficient operation.

Software Techniques for Low-Power Design

It is generally more effective to begin monitoring the energy consumption as early as possible to access the potential risks of high energy consumption points during the implementation process. When the software is already implemented and integrated, it is usually more difficult and expensive to eliminate such issues. On the other hand, energy consumption levels are directly proportional to computational complexities and improving one will result in indirect improvement of the other. Therefore, it is a good idea to introduce several software development techniques to achieve low-power in embedded systems.

Code optimization: Optimize algorithms to reduce the overall CPU utilization. Try using efficient algorithms and data structures to reduce the computational complexity. Frequently, there is a tradeoff between faster processing/larger code size vs slower processing/smaller code size. Usually, optimizing a code for speed vs size is a better choice.

Event-Based Task Scheduling: Events are generated to trigger the system to perform some work. Once the processor finishes the requested task, it goes back to idle state allowing it to remain in low-power modes for longer durations. Incorporating sleep modes in the code putsthe processor or specific peripherals into low-power states during periods of inactivity. Use of efficient task scheduling algorithms minimizes wake-up times and ensures that tasks are executed in a power-efficient manner.

Optimized Data and I/O Access: Minimizing unnecessary data transfers and using efficient data structures to reduce power consumption during memory access operation such as unnecessary copying of data, especially when large blocks of memory are allocated. Reducing the frequency of I/O operations and using techniques such as batch processing to minimize power consumption during data transfers. Optimizing cache usage to minimize memory accesses and reduce power consumption associated with accessing external memory.

Code Profiling and Optimization: Profiling code to identify power-hungry sections and optimizing them to reduce power consumption without sacrificing performance is a major area for optimization. Additionally, compilers that optimize code for low-power execution can significantly reduce energy consumption by minimizing unnecessary operations and maximizing sleep modes utilization. Debugging tools that provide insights into power consumption behavior during development help identify and solve power inefficiencies early in the design process.

Using Low-Power Communication Protocols

The adoption of low-power communication protocols within embedded systems is paramount for achieving energy efficiency while maintaining reliable data transmission. This section aims to offer insights into energy-efficient communication standards and wireless protocols customized for low-power applications.

Wireless Protocols for Low-Power Design

Wireless communication is gaining popularity in embedded systems for its adaptability and scalability. However, without energy-efficient implementation, it can lead to considerable power consumption. Several wireless protocols, tailored for low-power applications, have emerged to address this concern, including:

• BLE is designed for low-power devices and applications with infrequent data transmission.
• NB-IoT technology is designed to provide low-power wide area network (LPWAN) connectivity for IoT devices. This means that NB-IoT devices have very low-power consumption compared to traditional cellular devices, which enables them to operate on a single battery charge for years.
• Z-Wave is a highly efficient and low-energy technology. While the smart home hub requires a constant power supply to keep the network up and running, many Z-Wave devices operate on battery power alone for a year or more before requiring replacement.
• LoRa is ideal for IoT applications requiring low data rate transmission over long distances.
• ZigBee is a low-power, low-data-rate wireless communication protocol commonly used in home automation and industrial control systems.

Conclusion

Tauro Technologies can dramatically reduce system cost, size, and power requirements through optimized hardware and software design, and meticulous component selection. Our diverse portfolio of high-efficiency modules and integrated systems is engineered to meet the most demanding industrial standards. Contact us to explore how we can enhance your systems.

The post Embedded Systems and Low-Power Design appeared first on Tauro Technologies.

Wireless Communication in Embedded Systems

Embedded systems are extensively used in wireless and mobile communication systems, from smartphones and laptops to home appliances, industrial automation, and the Internet of Things (IoT). They leverage a variety of wireless communication protocols such as WiFi, BLE, Zigbee, Cellular, and Z-Wave. Collectively, these wireless protocols have brought us closer together and more connected than ever before.

In this blog post, we’ll delve into the essential elements of wireless communication technologies in embedded systems. We’ll explore the differences between low-power and high-power solutions, weigh the strengths of Bluetooth, Wifi, LoRa, Cellular Networks, Z-Wave, and Zigbee, look at the contrasts between long-range and short-range communications, and consider the trade-offs of high and low bandwidth. Furthermore, we’ll talk about communication architectures, differentiating point-to-point communication from mesh networks. We’ll also discuss how to cleverly blend these features to tailor them to the precise needs of your embedded application. So, let’s get started and unravel the world of wireless communication in embedded systems.

Wireless Communication Protocols in Embedded Systems

Embedded systems are like specialized multitaskers. They are built around tiny but powerful computers (microcontrollers, microprocessors) and are programmed to handle specific jobs. To get their work done, these systems need to talk to other devices, whether it’s through wired or wireless connections. This is where communication protocols come into play, which define how data is transferred between devices.

Wireless communication protocols are the backbone of seamless connectivity and data exchange in embedded systems, serving various applications, from IoT devices to industrial automation. Among the popular choices, Bluetooth Low Energy (BLE) shines for low-power, short-range communication, making it ideal for wearables and smart home appliances. Similarly, Zigbee finds its place in applications requiring low data rates, minimal power consumption, and short-range connectivity, often used in industrial control and home automation systems.

Choosing the right protocol depends on application-specific factors such as power efficiency, communication range, data rate, and network architecture. Choice is key to building robust, reliable, and scalable solutions that meet the diverse needs of an expanding embedded systems landscape. These protocols work together to build intelligent and connected automotive embedded ecosystems, enabling features such as real-time traffic updates, and in-car entertainment.

Radio Frequency (RF) transceivers serve as the cornerstone of wireless communications within embedded systems. These devices combine both transmission and reception functions, enabling a two-way flow of data over the airwaves. RF transceivers are versatile, facilitating communication in various protocols and frequency bands like Bluetooth, WiFi, Zigbee, and more.

Now, let’s take a look at some of the common communication protocols and technologies you’ll find in embedded systems:

• Bluetooth

Bluetooth is a short-range wireless technology standard that is used for data exchange between devices over short distances. It uses UHF radio waves of frequency ranging from 2.4 to 2.485 GHz in the ISM (industrial, scientific, and medical) radio band. In the most widely used mode, transmission power is limited to 2.5 milliwatts, giving it a very short range of up to 10 meters. Data can be shared at a maximum data rate of 3 Mbps.

Bluetooth is mainly used as an alternative to wired connections to exchange files between nearby portable devices. That is why Bluetooth Low Energy (BLE) and Bluetooth Classic radios are designed to meet the unique needs of developers worldwide.

– Bluetooth Classic is the original version of Bluetooth technology, which was designed for high-bandwidth applications. Operating over 79 channels in the 2.4 GHz ISM (Industrial, Scientific, and Medical) frequency band, it enables devices like phones and headphones to form personal area networks (PANs) to transmit data over short distances. Bluetooth Classic has become important to daily life, particularly as the trend toward smartphone devices without headphone jacks continues. The process of connecting two devices via Classic Bluetooth is now a common skill.

– Bluetooth Low Energy (BLE) is a version of Bluetooth technology designed for very low-power operation. Transmitting data over 40 channels in the 2.4 GHz ISM frequency band, this version provides developers a tremendous amount of flexibility to build products that meet the unique connectivity requirements of their market. BLE devices can run on a coin cell battery for months or even years. Although originally known for its device communication capabilities, BLE is now also widely used as a device positioning technology to address the growing demand for high-precision indoor location services. It now includes features that allow one device to determine the presence, distance, and direction of another device.

• WiFi

WiFi (Wireless Fidelity) is the most popular IoT communication protocol for wireless local area networks that utilizes the IEEE 802.11 standard through 2.4 GHz UHF and 5 GHz ISM frequencies. 2.4 GHz WiFi can reach a maximum of 600 Mbps in ideal conditions, but in an average home network, a max speed of 150 Mbps is more likely. A 5 GHz WiFi connection can reach up to 1300 Mbps. 2.4 GHz WiFi can reach up to 46 meters indoors and 92 meters outdoors, meanwhile, 5 GHz frequency spans around one-third of the distances of 2.4 GHz WiFi. It has a data rate of up to 600 Mbps maximum, depending on the channel frequency used and the number of antennas. In embedded systems, ESP series controllers from Espressif are popular for building IoT-based Applications.

There are many development boards available that allow people to build IoT applications using WiFi, for example, Raspberry Pi and Node MCU, which allow people to build IoT prototypes and also can be used for small real-time applications.

• LoRa

LoRa (Long Range) is a wireless technology that offers long-range, low-power, and secure data transmission for M2M and IoT applications. LoRa is based on chirp spread spectrum modulation, which has low power characteristics and can be used for long-range communications.

LoRaWAN provides the ability to connect millions of devices with data rates ranging from 0.3 kbps to 50 kbps. The distance for LoRaWAN application is up to 5 km in urban areas and up to 15 km or more in rural areas. With hundreds of millions of devices connected to networks in more than 100 countries and growing, LoRa is creating a smarter planet.

• Cellular Networks

Embedded systems also leverage cellular networks for wireless communication. The final stage of connectivity is achieved by segmenting the comprehensive service area into several compact zones, each called a cell. This protocol is generally used for long-distance communications. The data of larger size and with higher speeds can be sent compared to other technologies. The fifth generation of cellular networks is 5G. Its frequencies are divided between the Sub-6 GHz range, which has been extended to cover potential new spectrum offerings from 410 MHz to 7125 MHz, and the mmWave range, which includes frequency bands from 24.25 GHz to 71 GHz. The trade-off for speed at mmWave frequencies is the limited range of about 600 meters, while Sub-6 GHz frequencies can cover up to 5 km.

• Z-Wave

Z-Wave, low-power RF, is a communication protocol designed for Home Automation products. Smart home products with Z-Wave inside work together, using just one app to connect and control your smart home from anywhere. While Z-Wave has a range of 100 meters in open air, building materials reduce that range, it is recommended to have a Z-Wave device roughly every 10 meters, or closer for maximum efficiency. The Z-Wave signal can hop roughly 180 meters, and Z-Wave networks can be linked together for even larger deployments. Each Z-Wave network can support up to 232 Z-Wave devices allowing you the flexibility to add as many devices as you’d like to make sure your Smart Home is working its hardest. The data packets are exchanged at data rates of 100 kbps maximum and the protocol operates at a frequency of 900 MHz in the ISM band.

• Zigbee

Zigbee, like Bluetooth but with a longer range, relies on a bridge to facilitate internet-bound data from devices, albeit with higher power consumption. Zigbee has a shorter range of about 10-20 meters indoors because it uses less power. This does dramatically increase battery life for Zigbee devices. The data rate to transfer data between communicated devices is around 250 Kbps. It has a large number of applications in technologies like M2M and IoT.

This wireless communication standard, operating on the IEEE 802.15.4 standard, specializes in serving IoT applications. It offers dependable communication with low data rates and power usage, making it an ideal choice for applications where prolonged battery life is crucial. Zigbee finds its niche in domains like home automation, industrial control, and smart energy management systems, catering to various IoT needs.

Point-To-Point vs Mesh Topology in IoT Networking

Point-to-Point Topology and Mesh Topology are two distinct network architectures employed in computer networking. These topologies differ significantly in terms of their structure, connectivity, scalability, fault tolerance, and implementation.

Mesh Topology is a type of networking where all nodes cooperate to distribute data amongst each other. Originally developed over 30 years ago for military applications, mesh networks are now commonly used for various applications, including home automation, smart HVAC control, and smart buildings. Industry standards that rely on mesh network Topology include Zigbee and Z-Wave.

Mesh Topology can be further categorized into two types: Full Mesh and Partial Mesh. In a Full Mesh Topology, every device has a direct link to every other device, creating a fully connected network. In a Partial Mesh Topology, only some devices have direct links to all other devices, while others have links to only a subset of devices.

One significant advantage of mesh Topology is that it has low transmit power and shorter links (<100 ft). This characteristic not only extends the battery life significantly but also facilitates efficient data movement across the network. The other advantage of mesh Topology is its ability to facilitate self-healing networks in the face of node failures. If one node goes down, alternative connections can be established, enhancing fault tolerance. The primary disadvantage of mesh Topology is that the range between two mesh nodes is quite limited, which means that you may have to add additional nodes into your network that aren’t strictly necessary.

While mesh networks are well-suited for connecting multiple devices, Point-to-Point Topology is employed to connect two specific endpoints. Point-to-Point Topology is a network configuration where two endpoints have a direct connection or link. Serial connections between two devices or a USB connection between a computer and a printer are examples of point-to-point connections.

The primary advantage of Point-to-Point Topology is its simplicity, as it involves a direct flow of data either unidirectional or bidirectional between two points. Point-to-Point networks are still relevant in certain applications such as SCADA systems, traffic data systems, or Point-to-Point broadcast systems like police or fire radios. However, they are less suitable for IoT applications, where connecting to multiple nodes is often more practical.

The Future of Wireless Communications in Embedded Systems

The future of wireless communication in embedded systems promises remarkable advancements and innovations. Here are some key aspects that will define the future landscape:

The integration of 5G networks is set to bring a transformative impact to embedded systems. 5G offers significantly higher data rates, lower latency, and the ability to connect massive numbers of devices, revolutionizing real-time applications, ultra-high-definition video streaming, autonomous systems, and large-scale IoT deployments.

Private 5G networks differ from public 5G networks in that they provide restricted access and utilize licensed or unlicensed wireless spectrum within a confined area, such as a manufacturing plant, port, airport, campus, or business park. This allows owners to tailor the network to specific needs and requirements.

The key differences between public and private 5G have to do with restricted access and isolation. Typically, the public 5G networks available through service providers offer equal access rights to all users, sometimes leading to degraded service performance. A higher service availability is fundamental to support always-on operations. 

A private 5G network offers greater control. Unlike public 5G, a private 5G network can be reconfigured to permit different levels of access when certain network activities are deemed more business-critical than others.

Open RAN is a revolutionary shift in mobile network design, allowing service providers to use components from different vendors. This open approach is guided by industry standards, enabling flexibility and innovation in the creation of mobile network equipment. In Open RAN, traditional components like remote radio heads and baseband units are replaced with disaggregated radio units, distributed units, and centralized units. These components can be virtualized or containerized, offering programmable, intelligent, and interoperable functions. The O-RAN Alliance, established in 2018, defines the standards for Open RAN, bringing together global stakeholders in telecommunications. Open RAN represents the future of wireless communication, emphasizing openness, flexibility, and collaboration among suppliers.

As the volume of data generated by embedded devices continues to surge, edge computing will play a pivotal role in optimizing wireless communication. By relocating computation and data processing closer to the network’s edge, embedded systems can reduce latency, enhance real-time decision-making, and alleviate bandwidth constraints.

The concept of mesh networking, where devices communicate with one another to form a network without relying on a centralized infrastructure, holds significant promise for embedded systems. Mesh networks provide increased reliability, scalability, and flexibility. They enable self-healing capabilities, allowing devices to reroute data and ensure continuous connectivity, even in cases of individual node failures.

Conclusion

Choosing the most suitable network connectivity option for your IoT project is a decision that hinges on your device requirements and the ultimate project goals. It’s a delicate balance, primarily involving trade-offs among power consumption, available bandwidth, and network coverage.

In specific scenarios, technologies like LoRa and LoRaWAN can prove to be fantastic choices. For other cases, options like WiFi or Ethernet might be clear and straightforward. Furthermore, modern cellular solutions, exemplified by innovations like Notecard, have brought global IoT connectivity into the realm of reality.

The IoT landscape continues to expand and evolve, offering a diverse range of connectivity options serving various applications. By carefully assessing your project’s unique demands and keeping the trade-offs in mind, you can make an informed choice that aligns perfectly with your IoT objectives.

Interested to know more? Get in touch with us for details.

 

The post Wireless Communication in Embedded Systems appeared first on Tauro Technologies.

It is common for the majority to get confused with the terms “Embedded firmware” and “Embedded software”. In this article, we will discuss differences and similarities between embedded software and firmware and offer examples to help the reader differentiate between those two.

We will kick things off by getting to know what an embedded system is and exploring its core components. From there, we’ll dive into the challenges that developers encounter, from the intricate world of clocking mechanisms to navigating the nuances of firmware and managing power efficiently.

What is an Embedded System?

An embedded system is a computer system with a specific function, composed of a microprocessor, memory, and various input/output peripherals. These systems are often found within larger mechanical or electronic assemblies, hence the term “embedded”.

Embedded systems come in various forms, with some being standalone devices, while others function as integral parts of a larger system.

These systems have a presence in a wide range of applications, including industrial machines, consumer electronics, agricultural and processing equipment, automobiles, medical devices, cameras, digital watches, household appliances, airplanes, vending machines, toys, and even modern mobile devices.

Embedded systems consist of hardware and software components. The hardware includes microprocessor or microcontroller, memory, input/output interfaces, timers, and a power supply. These components require software and firmware to bring them to life and function as a system.

Challenges in Embedded Systems

Embedded product developers grapple with a multitude of challenges as they strive to design and develop efficient and reliable embedded systems. Here, we’ll explore some of the key challenges:

Clocking Challenges

• Synchronization: Achieving precise clock synchronization across different components within an embedded system is crucial for seamless operation. Variations in clock timing can lead to synchronization issues and data errors.
• Low Power Clocking: Balancing the need for high-performance clock speeds with power efficiency is a constant challenge, especially in battery-operated devices.
• Clock Domain Crossing: Managing different clock domains within a single system can be complex and requires careful attention to avoid synchronization problems.

Power Management Challenges

• Energy Efficiency: Balancing performance and power consumption is critical, especially in battery-powered devices. Achieving optimal energy efficiency while maintaining functionality is a constant struggle.
• Dynamic Power Management: Efficiently managing power in dynamic workloads, where system components operate at varying levels of activity, is a complex task.
• Thermal Management: Preventing overheating and thermal issues in embedded systems, which can affect performance and longevity, is another challenge.

Firmware Challenges

• Complexity: Developing firmware that is robust, efficient, and adaptable can be a significant challenge. Firmware must handle various tasks, from hardware control to communication protocols.
• Security: Ensuring the security of embedded systems is paramount. Firmware vulnerabilities can expose systems to cyber threats, making robust security measures essential.
• Compatibility: Firmware must often interact with diverse hardware components, requiring compatibility testing and updates as hardware evolves.

GUI and Dashboards in Embedded Systems

Graphical User Interfaces (GUIs) and dashboards play a crucial role in embedded systems, as they provide an interactive and user-friendly way to control and monitor devices and systems with limited computing resources.

What is Embedded Software?

Embedded software is designed to operate in SWaP-optimized non-PC devices. This software is designed for the specific hardware it runs on and often faces some problems due to limited processing power and memory capacity of the device.

A simple example of embedded software can be a controlling of household lighting using an 8-bit microcontroller with minimal memory. It can also be as complex as the software powering modern smart cars. These complex systems manage various electronic components, such as climate control, adaptive cruise control, collision detection, and navigation.

Embedded software and application software differ primarily in their scope and functionality. Embedded software is often serving as the device’s operating system itself. It operates under strict limitations imposed by the device’s functionality, which tightly controls the updates and additions to ensure compatibility.

On the other hand, application software provides specific functionality within a general-purpose computer and operates on a complete OS. This separation means that application software has more flexibility and fewer restrictions when it comes to utilizing system resources.

What is Embedded Firmware?

Firmware serves as a link between the hardware and other software applications that power the system. It is a special type of embedded software that was historically written in read-only memory (ROM) or electrically erasable programmable read-only memory (EEPROM). These earlier forms of firmware were notably unchangeable after initial programming. That is why it is called “firm”.

However, technology has evolved and moved toward storing firmware in Flash memory devices. This advancement offers notable advantages, including easier reprogramming and upgrade capabilities as well as significantly increased storage capacity when compared to its ROM and EEPROM predecessors.

Summing up, the primary role of firmware is to initiate device’s startup process and provide the essential orchestration to support the operation among various hardware components.

Hardware developers use embedded firmware for controlling hardware devices and their functionality similar to the way OS controls the function of software applications. Embedded firmware exists in everything from simple appliances that have computer control, like toasters, to complex tracking systems in missiles. The toaster would likely never need updating but the tracking system sometimes does.

The key difference between Embedded Software vs Firmware

Firmware is just a specific subset of embedded software. Without the operating system and middleware parts, firmware acts as a directional translator only and cannot work without other software layers working on top of it. It is just one layer, whereas a full embedded layer stack is required for a device to function.

Unlike the application software which is updated often, firmware is typically not updated after it is released and  working properly.

If we use a traffic light system analog here is how the embedded system components fit – Hardware (red) is the most difficult to update on a working product, firmware (orange) is not impossible but comes with challenges, and software (green) is easy to update and something that is being updated frequently.

Interested to know more? Get in touch with us for details.

The post Software and Firmware for Embedded Systems appeared first on Tauro Technologies.

As the demand for artificial intelligence continues to rise in various industries, from healthcare and finance to manufacturing and autonomous vehicles, industrial computers face the challenge of optimizing AI workloads. Developers are constantly seeking efficient and scalable solutions to solve these challenges. One such solution is using COM Express , a standardized form factor that can be used as a flexible computing platform for various AI workloads.

With the ability to choose from wide variety of CPUs and the flexibility to right-size CPU to target various AI workloads, COM Express empowers organizations to create efficient, scalable, and cost-effective AI solutions. In addition to harnessing the advantages of COM Express, developers can leverage additional AI accelerators to further optimize the solutions. COM-HPC, a new specification, further enables enhanced performance and scalability for high-performance computing applications.

The Intel Alder Lake x86 CPU is an ideal solution for COM Express modules targeting AI workloads due to built-in AI acceleration with Intel Deep Learning Boost technology. This integrated AI capability allows for efficient execution of AI workloads, such as neural network inference and deep learning tasks. By leveraging the built-in AI accelerator, COM Express modules based on Alder Lake can provide optimized performance for AI applications without the need for additional external accelerators.

What is COM Express?

COM Express is a highly integrated and compact computer on module that is designed to offer scalability and flexibility by providing a standardized form factor and interface for integrating different processor architectures and I/O configurations. Introduced by the PCI Industrial Computer Manufacturers Group in 2005, COM Express provides a single circuit board with integrated RAM.

This family of modular, small form factor modules has gained significant traction in various industries, including automation, gaming, retail, transportation, robotics, and medical fields. With eight different types, four sizes, and three major revisions, COM Express promotes vendor technology reuse while catering to mid-range edge processing and networking requirements.

The key differentiator of COM Express from traditional single-board computers (SBCs) lies in its ability to plug off-the-shelf modules into custom carrier boards designed for specific applications. This enables an upgrade path for the CPUs while keeping the carrier board intact. By using a custom COM Express carrier board, all necessary signals can be efficiently routed to the peripherals, while COM Express processor modules serve as the main controller. These advanced features ensure the versatility and adaptability of COM Express for diverse application requirements.

Comparing COM-HPC with COM Express

COM-HPC is an evolution of the COM Express standard, uniquely tailored to address the demands of high-performance computing applications. With its focus on enhanced performance, scalability, and advanced features, COM-HPC caters to the same applications and markets as COM Express, but with notable differentiators. It boasts higher-end CPUs, expanded memory capacity, and increased and faster I/O capabilities. It’s essential to emphasize that COM-HPC does not aim to replace COM Express, rather the two standards exist as distinct entities in the field of embedded computing, offering developers a broader spectrum of choices to meet specific application requirements.

COM-HPC brings significant improvements over COM Express for AI workloads, particularly in terms of PCIe lanes and PCIe generation support:

• Increased PCIe Lanes: One of the key advantages of COM-HPC over COM Express is the availability of more PCIe lanes. COM Express has a limited number of PCIe lanes, which can restrict the connectivity options and the number of I/O interfaces or accelerators that can be integrated. In contrast, COM-HPC modules provide a higher number of PCIe lanes, allowing for more extensive connectivity and the integration of multiple high-speed devices.
• PCIe Gen4/5 Support: Another crucial enhancement in COM-HPC is the support for PCIe Gen4 and Gen5, whereas COM Express supports up to PCIe Gen3. PCIe Gen4 and Gen5 offer higher data transfer rates and improved bandwidth compared to Gen3. This is particularly advantageous for AI workloads that require fast data movement between the CPU, GPU, storage devices, and other peripherals.

In summary, newer generation processors, paired with higher data rates, dramatically lower the size, power and cost requirements of the systems required to perform the AI tasks.

The Advantages of Choosing COM Express for AI Workloads

COM Express offers several distinct advantages when it comes to AI workloads. As a flexible and scalable platform, it provides developers to adapt their AI systems according to specific requirements like CPU performance, power requirements. They can then design a carrier board that integrates the module with additional AI-specific components, such as AI accelerators. Below is the block diagram example of COM Express platform with AI Accelerator.

Here are the key advantages of choosing COM Express (or COM-HPC) for AI workloads:

• Flexibility and Scalability: COM Express allows developers to choose from a wide range of CPU options. Such kind of flexibility allows them to choose the module that best matches the computing needs of their AI workloads. Whether it’s a complex neural network inference or deep learning task, the platform can be customized to deliver optimal performance.
• Modular Design: COM Express follows a modular design approach with a separate CPU module and carrier board. This modularity simplifies system customization and future upgrades. Developers can easily swap out or upgrade the CPU module without redesigning the entire system, saving time and effort while adapting to evolving AI requirements.
• Streamlined Integration: COM Express adheres to industry-standard form factors and interfaces, ensuring compatibility across different vendors. This standardized approach simplifies system integration, reducing development complexity and time to market. Developers can focus on optimizing their AI algorithms and software, confident that the hardware integration will be seamless.
• Rich Connectivity Options: COM Express provides a wide array of interfaces, including Ethernet, USB, PCIe, and DisplayPort interfaces. These interfaces enable effortless integration with various peripherals, sensors, and external devices commonly used in AI applications. The rich connectivity options enhance data I/O capabilities, facilitating efficient communication and interaction within the AI system.
• Long-Term Availability and Support: COM Express offers long-term availability and support, ensuring continuity for AI deployments. This is particularly crucial for industries that rely on stable and long-lasting AI systems. With a consistent platform and extended availability, developers can plan for long-term deployment and maintenance, with access to software updates and technical assistance.
• Cost Optimization: COM Express provides a cost-effective solution for AI workloads. By leveraging COM Express, developers can save on development costs and reduce time to market. The modular design allows for efficient resource allocation, ensuring optimal performance while minimizing unnecessary expenses.
• Time to Market: Since the computer modules are widely available in the embedded marketplace, COM Express enables developers to focus on the IO needs, the addition of accelerators, the AI models and application software.

Real-World Applications of COM Express for AI Workloads

As stated above, COM Express modules offer immense potential for developers to optimize AI workloads on industrial computers, leading to transformative impacts and various implications for cost-effective solutions and large-scale deployments. Let’s delve into real-world examples and insights to showcase the significance of this optimization trend.

In the field of autonomous vehicles, this optimization trend allows autonomous vehicles to navigate complex environments, enhancing safety and efficiency. By leveraging COM Express modules, developers can achieve cost-effective solutions by utilizing existing industrial computers and upgrading them with optimized AI capabilities, resulting in large-scale deployments of autonomous vehicles across transportation networks.

Industrial automation is another area where COM Express systems can revolutionize AI workloads. By optimizing AI algorithms on industrial computers using COM Express modules, developers can achieve significant cost savings and efficiency gains in manufacturing processes. For instance, AI-powered computer vision systems can inspect and detect defects in real-time, improving quality control and reducing production costs. The use of COM Express modules enables industrial computers to handle these AI workloads effectively, making cost-effective solutions viable for large-scale deployment in manufacturing facilities.

In the healthcare sector, COM Express systems can optimize AI workloads on industrial computers to improve diagnostics, patient monitoring, and personalized treatment. For example, by leveraging COM Express systems, developers can enable industrial computers to process complex medical imaging data and apply AI algorithms for more accurate and timely diagnosis. This optimization trend in AI workloads allows healthcare providers to deliver cost-effective, benefiting patients globally.

What to choose

AI accelerators are paired with COM Express module on the carrier as separate modules or integrated directly into the carrier board’s design. This modular approach provides scalability and flexibility, allowing system designers to customize AI processing capabilities to meet the specific requirements of their applications. It also enables easy upgrades or replacements of AI accelerators without having to modify the entire system, making it both cost-effective and future-proof. AI accelerators such as Blaize, Hailo or Axelera paired with COM Express module can provide significant benefits. For example, combining Axelera M.2 AI Edge accelerator module with COM Express Carrier board can achieve up to 120 TOPS of AI performance with the flexibility of switching between the CPU families for optimized compute needs.

These accelerators are specifically designed to enhance AI workloads and provide optimized compute capabilities compared to GPUs. This level of compute power can greatly benefit vision processing applications, which often require intensive computations for tasks such as object detection and classification.

Conclusion

COM Express and COM-HPC offer flexible and scalable platform to enable various AI workloads, allowing developers to customize their systems based on CPU performance, power requirements, and I/O interfaces. CPUs like Intel Alder Lake integrated into COM Express modules provide efficient AI execution, integrated graphics performance, enhanced compute density, ecosystem support, and broad connectivity options. The combination of the CPU with optional AI Accelerator delivers optimized performance, reducing costs and enabling efficient large-scale AI deployments.

With the Tauro Technologies’ team of electronic engineers and designers it becomes possible to design and deploy comprehensive AI processing systems based on x86 and ARM CPUs paired with various AI Accelerators. This strategic approach helps bring down costs and ensures the right balance between compute power and AI processing needed for the system. We can customize the I/O as well as the footprint to fit your application requirements.

Interested to know more? Get in touch with us for details.

The post Leveraging COM Express and COM-HPC for AI Workloads appeared first on Tauro Technologies.

As technology evolves, the automotive industry is constantly seeking ways to make driving safe, reliable, and autonomous. In this blog post, we’ll explore the features, functionality, and the impact that a platform based on dual NVIDIA’s AGX Orin modules offers for the future of vehicle safety during operation. Additionally, we will elaborate on the concept of safety-critical systems and highlight the distinctions between safety-critical functionalities and ADAS (Advanced Driver Assistance System).

The Jetson AGX Orin is designed for advanced robotics and AI edge applications for manufacturing, logistics, retail, service, agriculture, smart city, healthcare, and life science. Dual Orin (2 Orin devices on the same motherboard) offers system redundancy, which refers to the presence of backup or duplicate components that can take over in the event of a failure in the primary system.

ADAS provides driver assistance and convenience, but it is not solely responsible for critical functions that impact safety. Safety-critical functions encompass components directly involved in critical functions such as braking and collision avoidance. Safety-critical systems follow strict standards to ensure reliable operation.

What is Orin?

The NVIDIA Jetson Orin solution is a SOM (system-on-module) with CPU, GPU, memory, power management, and various high-speed interfaces embedded on a single board. NVIDIA Jetson brings accelerated AI performance to the edge in a power-efficient and compact form factor. The Jetson family of modules all use the same NVIDIA CUDA-X™ software, and support cloud-native technologies like containerization and orchestration to build, deploy, and manage AI at the edge.

NVIDIA’s Orin platform (SoC) has three series for its Jetson products:

• Jetson AGX Orin series
• Jetson Orin NX series
• Jetson Orin Nano series

NVIDIA Jetson Orin modules provide 275 TOPS of AI performance and which increases the performance 8 times compared to Jetson Xavier for multiple concurrent AI inference pipelines, in addition to high-speed interface support for multiple sensors.

One of the major features of NVIDIA Jetson Orin is the DLA (Deep Learning Accelerator) which supports next-generation NVDLA 2.0 with 9x the performance of NVDLA 1.0. It enables the GPU to run more complex networks and dynamic tasks.

A Comparison of Orin with Traditional CPU/GPU

Now, let’s delve into a comparison between traditional processors and Orin by examining the following key features:

Architecture

NVIDIA Jetson Orin is designed specifically for autonomous machines and edge computing. Jetson AGX Orin modules feature the NVIDIA Orin SoC with a NVIDIA Ampere architecture GPU, Arm® Cortex®-A78AE CPU, next-generation deep learning and vision accelerators, and a video encoder and a video decoder making it highly optimized for tasks like computer vision, deep learning, and robotics.

Traditional CPUs (Central Processing Units) and GPUs (Graphics Processing Units) are more general-purpose processors designed for a wide range of computing tasks, including running operating systems, executing applications, and performing graphics rendering.

Power Efficiency

NVIDIA Jetson AGX Orin series modules are designed with a high-efficiency Power Management Integrated Circuit (PMIC), voltage regulators, and a power tree to optimize power efficiency. It strikes a balance between performance and energy consumption, allowing for longer battery life and reduced power requirements in embedded systems.

While traditional CPUs and GPUs can offer high computational power, they are generally more power-hungry compared to specialized SoCs like Jetson Orin. They are commonly found in desktops, servers, and workstations where power consumption is less constrained.

AI Performance

The NVIDIA Jetson AGX Orin series provides server class performance, delivering up to 275 TOPS of AI performance for powering and managing autonomous systems. Its high performance is ideal for tasks like object detection, image recognition, natural language processing, and autonomous navigation.

Traditional CPUs and GPUs can also handle AI workloads, but they do not provide the same level of performance or efficiency as AI-focused modules like Jetson Orin. GPUs, in particular, have been utilized for parallel processing in deep learning tasks, but they are less power-efficient compared to specialized AI chips.

In addition, the Jetson Orin modules are extremely compact, enabling the compute platform to have reduced size and weight – critical for autonomous robots and UAVs.

Software Ecosystem

NVIDIA Jetson Orin is part of NVIDIA’s Jetson platform, which offers a comprehensive software stack, including drivers, libraries, and frameworks specifically optimized for AI and autonomous applications. It supports popular AI frameworks like TensorFlow, PyTorch, and CUDA, providing developers with familiar tools and resources.

Traditional CPUs and GPUs also have a mature and extensive software ecosystem with support for a wide range of operating systems, development tools, and programming languages. They are compatible with various software frameworks, including those used for AI, but may require additional configuration and optimization for specific AI workloads.

Key differences between NVIDIA Orin and Xavier

NVIDIA Jetson AGX Xavier and NVIDIA Jetson AGX Orin have the same physical footprint and are pin compatible while also being in the same price range with one major difference that the Orin offers much higher performance.

The biggest change change is moving from Nvidia’s Carmel CPU clusters to the ARM Cortex-A78AE on Jeston AGX Orin.
The Orin CPU complex is made up of 12 2.2 GHz cores, each with 64KB Instruction L1 Cache and 64KB Data Cache, and 256 KB of L2 Cache. This enables x1.85 performance increased compared to the eight core Carmel CPU on Jetson AGX Xavier.

Jetson AGX Orin modules deliver an AI performance that can reach 275 TOPS with up to 64 GB of memory, compared to 32 TOPS with up to 32 GB of memory for Jetson Xavier.

Jetson AGX Orin 64GB has 2048 CUDA cores and 64 Tensor cores with up to 170 Sparse TOPS of INT8 Tensor compute, and up to 5.3 FP32 TFLOPs of CUDA compute, while Jetson Xavier has only up to 1.4 FP32 TFLOPs of CUDA compute. Ampere GPU brings support for sparsity, a fine-grained compute structure that doubles throughput and reduces memory usage.

DLA 2.0 provides a highly energy efficient architecture. With this new design, NVIDIA increased local buffering for even more efficiency and reduced DRAM bandwidth. DLA 2.0 additionally brings a set of new features including structured sparsity, depth wise convolution, and a hardware scheduler. This enables up to 105 INT8 Sparse TOPs total on Jetson AGX Orin DLAs compared with 11.4 INT8 Dense TOPS total on Jetson AGX Xavier DLAs.

The 12-core CPU on Jetson AGX Orin 64GB enables 1.85 times the performance compared to the 8-core NVIDIA Carmel CPU on Jetson AGX Xavier. Customers can use the enhanced capabilities of the Cortex-A78AE including the higher performance and enhanced cache to optimize their CPU implementations.

Jetson AGX Orin modules bring support for 1.5 times the memory bandwidth and 2 times the storage of Jetson AGX Xavier, enabling 32GB or 64GB of 256-bit LPDDR5 and 64 GB of eMMC. The DRAM supports a max clock speed of 3200 MHz, with 6400 Gbps per pin, enabling 204.8 GB/s of memory bandwidth.

The combination of NVIDIA’s processing capabilities and power efficiency, along with its safety-critical features, makes it the ideal solution for autonomous applications.

Safety Critical Software in Automotive Safety

Functional safety in processor-based systems is particularly critical in automotive applications. Apart from the ongoing shift towards autonomous vehicles, cars are increasingly dependent on microprocessors to carry out essential operations and must have redundant systems to enable safety in the event of a component failure.

ISO 26262 serves as the globally recognized standard for ensuring functional safety in the automotive industry. This international standard encompasses both the hardware and software components of a vehicle’s electrical and electronic (E/E) systems. Throughout the development process, ISO 26262 outlines specific requirements that must be fulfilled to ensure the safety-related functionality of the system, along with the corresponding processes, methodologies, and tools. By adhering to the ISO 26262 standard, manufacturers can ensure that sufficient safety measures are implemented and maintained throughout the entire lifespan of the vehicle.

ISO 26262 offers comprehensive guidelines on determining acceptable risk levels for systems or components and documenting the testing process. It encompasses the following key aspects:

• Defines an automotive safety lifecycle that covers management, development, production, operation, service, and decommissioning stages, allowing for customization of activities during each phase.
• Implements an automotive-specific risk-based approach for classifying risk levels known as Automotive Safety Integrity Levels (ASILs).
• Utilizes ASILs to specify the required safety measures for achieving an acceptable residual risk.
• Establishes requirements for validation and confirmation measures to ensure the attainment of a satisfactory level of safety.y

Dual AGX Orin Controller Overview

The Dual AGX Orin system offers superior computing power compared to a single Orin solution, making it preferable for specific applications that require higher computational power and redundancy.

The Dual Orin Controller’s computational capacity enables it to handle multiple complex tasks simultaneously. This capability is particularly valuable in scenarios where there is a need for concurrent processing of multiple data streams from various sensors, making it suitable for advanced autonomous machines, commercial vehicles, unmanned distribution vehicles, and unmanned cleaning vehicles.

In safety-critical applications, redundancy is essential to ensure system reliability. The Dual Orin Controller’s utilization of two AGX Orin modules provides a level of redundancy and failover capabilities. If one module encounters an issue, the other can continue functioning, minimizing the risk of critical system failures and improving the overall reliability of the autonomous machine.

Tauro Technologies TT300 Dual AGX Orin Controller

Tauro Technologies’ TT300 Dual AGX Orin compute platform provides exceptional computing power, low energy consumption, in a compact form factor.

With up to 400/550 TOPS of AI performance this product can be used in autonomous vehicles, UAVs and robotics. The product is designed for high reliability and redundancy, provides multi-sensor clock synchronization with sub-nanosecond accuracy and millisecond latency for precise timing.

Let’s take a closer look at TT300 key features:

Dual Orin Controllers 550 TOPS

The TT300 board is equipped with two powerful Orin controllers, delivering combined processing power of 550 TOPS. This immense computing power enables lightning-fast data processing and analysis, making it ideal for handling complex AI workloads.

Infineon TC397 Safety MCU

Ensuring the highest levels of safety and reliability, the TT300 board incorporates the Infineon TC397 safety microcontroller to support safety requirements up to ASIL-D. This MCU plays a crucial role in safeguarding the system against potential hazards and maintaining the integrity of critical operations.

100Base-T1/1000Base-T1 Ethernet

To facilitate efficient and reliable data communication, the TT300 board is equipped with both 100Base-T1 and 1000Base-T1 Ethernet interfaces. These interfaces enable fast and secure data transfer, ensuring smooth integration into existing vehicle network infrastructures.

Wi-Fi/4G/5G

TT300 board supports Wi-Fi, 4G LTE and 5G connectivity, enabling seamless wireless communication and remote access. Whether you need to stream data, receive updates, or control the board remotely, these connectivity features have you covered.

• GMSL2 Interface for Hi-Res Cameras

The TT300 board features a GMSL2 interface, enabling reliable connection with high-resolution cameras. This interface supports the transmission of data between the controller and cameras, ensuring high-quality image and video feed for AI applications such as ADAS, object detection, tracking, and recognition.

GMSL cameras are becoming a defacto standard in automotive industry where high data rates and long-distance support is required, addressing the need to transport higher video data rates in automotive video systems.
In addition to high bandwidth transmission, long-distance support, and low latency, GMSL cameras also come with the following features:

• Virtual channel support
• GMSL1 and GMSL2 backward compatibility
• Video duplication
• Automatic Repeat Request (ARQ) feature
• Compatibility with ARM platforms like the NVIDIA Jetson series

I/O Capabilities

TT300 is powered by two NVIDIA Jetson AGX Orin modules and Infineon TC397 safety MCU enables the design to meet ASIL-D highest reliability requirements. The I/O capabilities of the product include automotive as well as industrial ethernet interfaces, USB, wireless connectivity over 4G/5G and Wi-Fi, GMSL camera and LVDS radar interfaces for ADAS applications, as well as CAN and LIN interfaces for automotive and robotics applications routed to CMC connector. Wide selection of interfaces and customization options makes this device easily adaptable to various use cases and application scenarios.

Conclusion

Tauro Technologies’ TT300 is one of the industry’s first platforms to offer the NVIDIA Jetson Orin AGX in a redundant safety-critical setting. This is an ideal system for self-driving vehicles in automotive, mining, and defense sectors as well as autonomous robots and UAVs that require exceptional performance and functional safety certification.
We can customize the I/O as well as the product packaging to fit your application requirements – contact us for details.

The post Dual Orin Controller: The Ideal Safety-Critical Platform for Autonomous Vehicles appeared first on Tauro Technologies.