Which Interface is Best for High-Throughput Domain Controllers?

Summary

The automotive industry’s accelerating move toward centralized domain and zonal electronic/electrical (E/E) architectures indicates there should be a re-thinking of the communication interfaces for high throughput sensors especially Inertial Measurement Units (IMUs). In this paper, we examine the suitability of four common interfaces, including SPI, I2C, UART, and CAN FD, for the interconnection of IMUs with high-performance domain controllers in the vehicles with an emphasis on data throughput and real-time performance¹.

• SPI (Serial Peripheral Interface) is seen as the best option for high frequency low latency data acquisition from a “raw” IMU placed on the same printed circuit board (PCB) as the domain controller’s main processor. By having a powerful bandwidth (<100 Mbps) with very limited protocol overhead it is perfect for close coupled and real-time sensor fusion problems².
• CAN FD (Controller Area Network with Flexible Data-Rate) is the way which shines through for connecting physically remote or “smart” IMU modules to a domain controller. With its robust, differential-signal capability, built-in error-handling, and ample bandwidth (up to 8Mbps) are the reliability that in-vehicle networking reflects³.
• I2C and UART have generally been found to be insufficient for the major high rate stream of data emitted by contemporary IMUs because of the higher overheads and bandwidth limitations. Their role is limited and they have values for ancillary functions such as device configuration or low frequency diagnostic reporting.

1. Introduction: The Architectural Evolution That Is Changing the Requirements of Interfaces

The automotive industry is in the middle of a fundamental paradigm shift from a distributed architecture with dozens or even hundreds of individual Electronic Control Units (ECUs) towards a centralized architecture consisting of a few powerful domain controllers^{4 5}. These domain controllers, which are in charge of things such as Advanced Driver-Assistance Systems (ADAS), infotainment or vehicle motion, must receive, process, and fuse reams of data from a plethora of different sensors, including cameras, radar, LiDAR and IMUs^{6 7}.

This architectural consolidation is enforced by the need to deal with complexity, wiring and cost, and software-defined vehicles^{8 9}. However, it places tremendous stress on the communication interfaces that are responsible for ferrying data from the sensors to the central processing units. The IMU in particular is a key part of ADAS and autonomous driving capabilities, as it delivers useful information of theુર vehicle’s angular rate and linear acceleration to perform localization, plan the vehicle path, and control the vehicle stability. High frequency data returns from today’s IMUs are easily capable of saturating even the most modern and fast bus interfaces, making the selection an important design choice¹⁰.

This report covers an in-depth, comparative analysis of four major communication interfaces (for high-throughput, real-time data transmission for IMU) namely SPI, I2C, UART, and CAN FD. The following analysis looks at each interface from the perspectives of its theoretical bandwidth, practical aspects of throughput, real-time properties such as latency and jitter and its strategic position in emerging vehicle E/E topologies.

2. Comparing Mode of Communication Interfaces

The choice of an interface depends on a compromise between speed, robustness, distance, complexity and cost. Each of the following protocols stands alone in the multidimensional landscape in its own unique position.

2.1. SPI (Serial Peripheral Interface)

The SPI is a synchronous, full dual, master-slave serial protocol. It is well-known for its simplicity and high speed that make it a typical option to be used for on-board communication between microcontrollers and peripheral chips.

• Bandwidth and Throughput: SPI is always the fastest interface of the four contenders for short-distance communication. Theoretical clock speeds stretch out to tens of mega Hertz (MHz). For example, certain IMU chips have SPI clock frequencies up to 24MHz, which allows 24Mbps raw data rates^{11 12}. Its simple protocol structure, having no overhead of start/stop bits and no complicated addressing scheme as other protocols, means that its practical throughput is very close to its theoretical bandwidth . This makes it very efficient at streaming continuous high frequency data.
• Real-time Performance (Latency and Jitter): SPI offers extremely low latency because it is a synchronous communication and there does not appear to be any high protocol overhead. The delay between a request and the start of transmitting data is typically a matter of transmitting a few clock cycles. The search results lead to microsecond-level measurement of SPI transactions¹³. One source recorded the maximum jitter of a particular sensor over SPI at only 3.9μs¹⁴. This low and predictable latency is of key importance for the sensor fusion algorithms in which the time synchronization of multiple data streams is a critical requirement.
• Strengths and Weaknesses:
  • Strengths: Conveyance over speed very high, overhead protocol small, become to communicate in full duplex, a relatively simple hardware implementation¹⁵.
  • Weaknesses: Require additional signal lines (typically 4 – SCLK, MOSI, MISO and SS), no built-in mechanism for flow control and error detection, normally limited to short distances on the same PCB^{15 16}.
• Role In Domain Controller Architecture: SPI’s unique characteristics make it the ideal choice of connecting the “raw” IMU sensor chip directly in to the main System-on-a-Chip (SoC) or microcontroller from the domain controller module itself. This closely coupled connection guarantees the highest possible rate of communication for sensor data to the processing core for immediate fusion and processing.

2.2. I2C (Inter-Integrated Circuit)

I2C is a synchronous, half-duplex, multi-device serial bus which is known for its usage of only two wires (SDA and SCL) for communication.

• Bandwidth and Throughput: I2C is quite slow as compared to SPI. Standard modes run at 100 kbps and “Fast Mode” at 400 kbps while “High-Speed Mode” is up to 3.4 Mbps^{17 18}. However, its protocol overhead, start/stop conditions and device addressing and acknowledgment bits for each and every byte decrease the effective data throughput considerably¹⁹. This overhead makes it inefficient for streaming as large and continuous of data packets as a high rate IMU would.
• Real-time Performance (Latency and Jitter): The addressing and acknowledgment scheme inherent to the I2C protocol adds more latency and less determinism to the protocol than SPI. Bus arbitration, in which more than one master device may be contending for the bus, may further increase the latency unpredictably. While suitable for command and control tasks, this makes it less ideal for task requirements that need time-sensitive, high-frequency data streams for use in a real time control loop.
• Strengths and Weaknesses:
  • Strengths: Needs only two wires, can offer more than one device on the same bus and is easy in addressing scheme^{20 21}.
  • Weaknesses: Lower speed, greater protocol overhead, half duplex communications^{19 20}.
• Role in Domain Controller Architectures: Due to its bandwidth limitations, I2C is not a good choice for sending the main high frequency data from an IMU. Instead, it is best used for ancillary functions, such as the ability to configure internal registers on the IMU sensor at startup or to read out low frequency data such as temperature, or communicate with other low bandwidth peripherals on the domain controller board.

2.3. Universal Asynchronous Receiver/ Transmitter (UART)

UART is one of the simplest types of serial communication that is asynchronous and normally a point-to-point type.

• Bandwidth and Throughput: It is defined as baud rates (commonly 9600 bps or 115,200 bps) and they are the rate of data transfer in the network^{22 23}. While you may get much higher speeds than this, it is generally a low-speed interface. The asynchronous nature requires the use of start and stop bits for each data byte, and thus has a fixed overhead of at least 20%, and the effective throughput is therefore further reduced²¹.
• Real-time Performance (Latency and Jitter): Since the asynchronous nature and strict matching between transmitter and receiver are used, the timing error and jitter are common problems in using the UART. Its transmission character-based and lack of shared clock results in a greater and more variable latency than the synchronous protocols such as SPI.
• Strengths and Weaknesses:
  • Strengths: Very simple to implement, very available and only two wires are needed for simple communication (Tx, Rx)^{21 20}.
  • Weaknesses: Slow speed, high protocol overhead and usually limited to point-to-point connection²¹.
• Role in Domain Controller Architectures: In the context of high-performance data acquisition, it can be said that the role of UART is almost an obsolete interface. In a modern domain controller, its role is limited almost exclusively to debugging, console access, or communication with simple and low data rate legacy modules. It is completely unsuitable for high throughput IMU data.

2.4. Controller Area Network with Flexible Data-Rate (CAN FD)

• Bandwidth and Throughput: CAN FD brings two major enhancements; increased data payload up to 64 bytes in comparison to 8 bytes for classic CAN and flexible increased data rate for the data portion of the message frame²⁴. While the arbitration phase is still at the standard CAN speed (e.g., 500 kbps or 1 Mbps), in the data phase it can be fast enough to run at speeds of 2Mbps, 5Mbps or even 8Mbps in some implementations²⁵. This combination dramatically increases the effective data throughput and research has shown an effective increase in throughput of up to 10 times or more compared to classic CAN^{26 27}.
• Real time Performance (Latency and Jitter): As it is a Bus Protocol, CAN FD’s Latency is affected by the bus load. But the arbitration process, in which the messages are prioritized according to their identifier, guarantees the deterministic behavior but introduces latency and, in particular, for messages with low priority in a busy network. However, for a high-priority message stream such as IMU data the latency can be predictable and low. Its main advantage is its differential signaling and strong error detection and processing mechanisms which ensure message integrity even in the harsh electromagnetic atmosphere of a vehicle.
• Strengths and Weaknesses:
  • Strengths: Very robust and reliable, very good error handling, longer distances than SPI/I2C, great bandwidth improvement as compared to classic CAN.
  • Weaknesses: Point to point: Slower than SPI, Latency is dependent on Bus Traffic, Protocol is more complex than SPI.
• Role in Domain Controller Architectures: CAN FD is not intended to be used instead of SPI for on-board communication. Instead, it is particularly good as a network at the vehicle level to connect the intelligent sensors or subordinate ECUs with the domain controller. An IMU module which is away from the domain controller (i.e. located at the center of gravity of the vehicle for the best readings) would be a perfect candidate for a CAN FD interface. The appearance of IMU products with native CAN FD capability is a good example of this trend.

3. Tiered Interface Strategy in the Centralized Architectures

The emergence of the centralized domain controller has not resulted in the dominance of one interface. Instead, it has promoted a complex, tiered approach to communication, in which various interfaces are used depending on the best operating space.

• Tier 1: On-Board, Chip-to-Chip Communication: At the heart of the domain controller is data requiring transfer between the central SoC, Memory and critical peripherals on the same PCB with maximum speed and minimum latency. This is the domain of SPI. Its great bandwidth is necessary to feed the processing core with raw data from a co-located IMU, so that the sensor fusion algorithms have the far most immediate and untouched data as possible.
• Tier 2: In-Vehicle, Module-to-Controller Networking: When sensors are being distributed all over the vehicle, there is need for a robust network. This is where CAN FD shines. It offers a reliable connection with enough bandwidth for smart sensors to possibly take a few steps in filtering or processing data before sending the results to the domain controller. Being proven in automobile environments, it makes it a safe and effective method of connecting an off-board IMU module.
• Tier 3: The High-Bandwidth Backbone: For the communication between different domain controllers or for streaming extremely high-bandwidth data from sensors such as cameras and LiDAR, Automotive Ethernet is becoming the backbone of the vehicle network
  ^{28 29}. With speed ranges of 100Mbps up to multiple Gbps and protocols for time-sensitive data networking (TSN), Ethernet is ideal for managing the massive data aggregation that is needed by centralized fusion platforms. While direct IMU to Ethernet connections are only just starting to appear, direct connection of these will be the future for complex sensor pods that combine multiple sensor types.

This tiered approach can be seen to show that the question is not “SPI vs. CAN FD”, but can rather be read to mean “Where is SPI most appropriate and where is CAN FD most appropriate”.

4. Final Analysis: Choosing the best Interface for a high through rate IMU data

Based on the above analysis, the best interface between an IMU in a domain controller architecture is dependent upon its physical and logical position in the system.

• For a raw IMU sensor physically which is on the domain controller PCB: SPI is the indubitable better option. Its unrivalled speed and its near zero protocol overhead guarantees the lowest possible latency which is of the utmost importance for real-time application such as vehicle stability control and sensor fusion for autonomous driving. The processor receives the rawest data, with the smallest amount of delay, in order to make the most effective use of its fusion algorithms.
• For a smart IMU module located at a remote position in the vehicle: CAN FD is the most suitable and robust one. It is enough that the IMU module can do some preliminary processing and then send the filtered data packets (up to 64 bytes) via CAN FD bus. The interface supplies the bandwidth needed while assuring the high level of reliability and failure resistance for communication across vehicle wiring harnesses. This architecture is found on many production ADAS systems where the IMUs are positioned where they can provide the greatest performance .

Conclusion

The evolution towards centralised domain controllers has increased the need for high throughput / real-time capable interfaces. For connecting IMUs there is no one winner in this particular application, but rather a clear delineation of roles according to architectural requirements.

• SPI is still the king of on-board, fast and low-latency communication and therefore is the ideal interface with tightly-integrated IMUs connected to the central processor on the same board.
• CAN FD has earned itself as the modern standard of robust, medium to high speed in-vehicle networking and as such the best candidate to connect distributed or smart IMU to the domain controller.
• I2C and the Universal Asynchronous Receiver/Transmitter (UART) get put on the shelf with legacy and low planning tasks, not enough to be used with the data rates that modern IMUs throw out.

Therefore, expert system designers wouldn’t have one interface rather than the other but would carefully drive in both SPI and CAN FD in a tiered system architecture to take advantage of the unique benefits provided by each in order to achieve a high performance and robust system.

1. 熊莉. 计算机行业研究框架（汽车智能化、工业数字化）. P. 67, 国信证券经济研究所, 10 July 2022.
2. InvenSense, Inc. IAM-20380HT High Performance Automotive Grade2 3-axis Gyroscope. P. 20, Document Number: DS-000521, Revision: 1.0, 14 Sept. 2022.
3. Jung WooShin, Jung Hwan Oh, Sang Muk Lee, and Seung Eun Lee. “CAN FD Controller for In-Vehicle System.” 2016 International SoC Design Conference (ISOCC), 23–26 Oct. 2016, Jeju, Korea (South). IEEE, doi:10.1109/ISOCC.2016.7799870.
4. Greg Basich, Infotainment and Telematics: In-Vehicle Infotainment – The Automotive Cockpit of the Future. P. 5, Strategy Analytics, Dec. 2018.
5. Ron DiGiuseppe, Enabling Centralized ADAS Domain Controller SoCs with Automotive IP. P. 1, Synopsys, Dec. 2018.
6. Global and China Automotive Domain Control Unit (DCU) Industry Report, P. 3, 2018-2019. ResearchInChina, Feb. 2019.
7. ADAS Domain Controller Key Component Trends Report, 2022. ResearchInChina, Jan. 2023.
8. Domain Controller: a key component for the cars of tomorrow, Valeo
9. Ryan Suzuki, Samsung’s 12nm-Class Automotive LPDDR5X: DRAM for Safety-Critical Centralized Automotive Systems
10. M-G366PDG0 Full Data Sheet. Rev. 1.2, M-G366PDG0, SEIKO EPSON CORPORATION, 2023.
11. ICM-42607-P Datasheet: Premium Performance 6-Axis MotionTracking^TM IMU. P. 6, Document Number: DS-000417, Revision: 0.1, InvenSense Inc., 20 Oct. 2020.
12. ICM-42605 Datasheet: Premium Performance 6-Axis MotionTracking^TM IMU. Document Number: DS-000292, Revision: 1.7, InvenSense Inc., 12 Sept. 2022.
13. Alex Grudzinski, and Dylan O’Brien. Designing Smarter Automotive Systems with MSPM0 AEC-Q100 MCUs. P. 23, Texas Instruments, Webinar, 2023.
14. AIS1120SX / AIS2120SX Datasheet: MEMS Automotive Acceleration Sensor: Single/Dual-Axis with SPI Interface. DocID028312 Rev 4, STMicroelectronics, Apr. 2020.
15. STM32 – UART、IIC、SPI通信协议 
16. UART、I2C、SPI、Ethernet, 汽车ECU开发
17. 机器人、无人机是怎么知道自己的姿态的？揭秘 IMU 的关键参数, EETOP, 2025
18. Javier de Esteban, Integración de sensores IMU para la navegación autónoma de robot. P. 21, Directed by Alba Perez Gracia, Escola Tècnica Superior d’Enginyeria Industrial de Barcelona, Quadrimestre primavera 2023.
19. Lothar Thiele and Michele Magno. Embedded Systems: 1 – Introduction., P. 132, ETH Zurich, Computer Engineering and Networks Laboratory, D-ITET Center for Project-Based Learning, 2022.
20. 常见通信的底层数据传输原理（UART、SPI、I2C）, STM32嵌入式开发, 2024
21. UART、I2C、SPI、Ethernet, 汽车ECU开发, 2022
22. spi、iic、can高速传输速度与选择
23. Datsiuk, D., et al. “Orientation and Stabilization of Marine Unmanned Systems with IMU: An Experimental STM32-Based Approach.” Sciences of Europe, no. 170, 2025, pp. 45-46. DOI: 10.5281/zenodo.16803338.
24. Robert Hoettger, et al. Specification of In-car Software Architecture for Car2X Applications. P. 62, APPSTACLE Project Deliverable D1.1, ITEA 3 – 15017, 29 Jan. 2018.
25. 驱动汽车电子创新：RIGOL CAN-FD总线分析解决方案, 电子产品世界
26. Andres Zavala, and Arturo Inzunza. Advanced Vehicle Networking: AMF-AUT-T0502. P. 8, Freescale, September 2013.
27. Harald K. Eisele, and Natalie A. Wienckowski. CAN FD: From Concept to Production. CAN FD Symposium, 16 Feb. 2017.
28. Camille Kokozaki, Designing Integrated ADAS Domain Controller SoCs with ISO 26262 Certified IP, 2018
29. Automotive Ethernet Testing Market by Test Type (Conformance Testing, Electromagnetic Compatibility Testing, Environmental And Stress Testing), Test Level (Component Level, Domain Controller Level, ECU Level), Test Equipment Type, Service Type, Vehicle Type, Application, End User – Global Forecast 2025-2032, Oct. 2025