iCC 2013 CAN in Automation

The Potential of CAN FD Technology to Impact Upon FlexRay

C. Quigley, A. Williams, R. McLaughlin, Warwick Control Technologies Ltd.

Abstract - CAN FD has been introduced as a way of improving the data throughput and integrity of CAN bus systems. FlexRay is the emerging technology for automotive high speed control networking. The adoption of a faster network such as FlexRay has the ability to reduce the weight of the vehicle electrical architecture by replacing a number of lower speed networks with a single faster network. This in turn reduces the number of gateway ECUs, wiring and connectors, and reduces system complexity. By increasing the data throughput CAN FD has the potential for car manufacturers to delay the adoption of a faster network such as FlexRay by breathing new life into existing CAN architectures. In this paper the features of CAN FD and FlexRay are compared in terms of the cost of implementation, data throughput and protocol complexity. A conclusion is made on the impact of CAN FD on FlexRay technology.

Introduction

CAN FD has been introduced as an enhancement to the CAN protocol by providing improved bandwidth [1]. The FlexRay protocol has enjoyed some adoption in high-end cars as a high speed network. The aim of this paper is to look at the potential impact of CAN FD on FlexRay technology by comparing some key aspects of these technologies. Therefore this paper has an automotive perspective. The key features of CAN FD and FlexRay are compared at a top level. Then some analysis is carried out on the potential bandwidth of both technologies and a comparison between the two is made. Finally CAN FD and FlexRay are compared in terms of cost and application to automotive high integrity systems.

High-Speed Networking Benefits

FlexRay has become a successful technology for high speed automotive control networks. It was originally aimed at providing features for X-by-wire systems. However the biggest benefit has Figure 1: Example complicated been on a system design perspective in architecture that can be simplified using which a higher speed network such as a high-speed networking technology FlexRay can provide a number of benefits.

02-19 iCC 2013 CAN in Automation

Firstly a single faster network can replace luxury brand Rolls Royce adoption into several lower speed networks. This in turn these platforms is likely. Wider adoption reduces the number of gateway Electronic in the automotive industry is currently Control Units (ECUs), reduces the amount unclear. A number of research projects for of wiring in the vehicle due to the reduction FlexRay have taken place in aerospace in the number of networks which therefore and robotic control applications. reduces the weight of the electrical The implementation of FlexRay for chassis system. Secondly the deletion of one or control in a production BMW was more networks reduces the control system described by Berwanger et al in 2005 [2]. complexity and less effort needs to be Although the nodal costs of FlexRay are taken to decide on how to partition a higher, an interesting reason for choosing network and how to transport a signal FlexRay stated by Berwanger et al is that it which may need to exist on several CAN can reduce system costs when compared buses [2]. For example in figure 1 there to a CAN implementation. This is counter- are three CAN buses each of which must intuitive when nodal costing is considered. use the wheel speed signals. In this However, it is argued that cost savings can example these signals originate on the be enjoyed by replacing several CAN chassis CAN bus but would also be used buses. This in turn reduces wiring, on the powertrain CAN bus by the engine connectors, gateway ECUs and all of this and gearbox controllers and be processed reduces system partitioning effort. They in the instrument cluster to display vehicle also stated that reducing the number of speed. It would most likely also be used CAN sub-buses, cables and redundant on the Infotainment bus for the in-car radio sensors meant that, holistically, the for speed adaptive volume control. implementation of FlexRay is roughly the same as for CAN. They also estimate that Summary of Known FlexRay Adoption integration of the FlexRay controller into the microcontroller and using a lower cost The FlexRay consortium was launched in transceiver will save approximately three 2000 to develop the FlexRay specifications and one Euros respectively per ECU. and market. It consisted of core members However, no real data has been presented BMW, DaimlerChrysler, Freescale to support these arguments and also it is (previously called Motorola), Philips not clear whether this considers the impact Semiconductors, Robert Bosch and of adopting a new and sophisticated Decomsys. The first draft of the FlexRay technology such as FlexRay. protocol was introduced in 2001. The aim of FlexRay is to complement CAN in Motivation for CAN FD higher bandwidth and integrity applications and is now at revision 3.01. This particular revision has been announced for inclusion FlexRay is a way of providing more in the latest BMW X5 chassis control bandwidth and as previously described system [2]. has already been adopted on a small FlexRay has established a significant number of high-end vehicles. However advance on CAN technology by increasing there is at least one use case in which bit rate, providing synchronisation between FlexRay is not the best solution. FlexRay nodes so that a time triggered bus access is not an efficient protocol for ECU methodology is achieved and providing a flashing. This is because the protocol backup data channel for dual mode would require that an ECU undergoing a redundancy. flashing operation must wait for Since 2005 FlexRay has enjoyed some appropriate spare time in the adoption in some high-end automotive communication schedule to be able to such as several BMW platforms and also handle flash data. Traditional CAN extended into the Audi A8 initially before connected via the OBD is very slow as it extending into other Audi and Bentley runs at 500 Kbit/s and FlexRay can be fast vehicles. Since the BMW group own the running at 10Mbit/s but is not flexible

02-20 iCC 2013 CAN in Automation enough due to its time-triggered To help keep the comparison consistent communication. Therefore CAN FD has between the three technologies it was been proposed initially to at least close the assumed that the DLC was 8 bytes. The gap between CAN (max. 1 MBit/s) and comparison is shown in Figure 2SEQ FlexRay (10 MBit/s) for ECU flashing type which shows maximum data throughput in applications. bytes per second versus the bit rate in the Another motivation for CAN FD is the high data phase. Overlaid on this diagram is effort for migration for an automotive OEM CAN at 500 Kbit/s and FlexRay at and its suppliers to move to a faster 2.5Mbit/s. FlexRay has three possible bit technology such as FlexRay or Ethernet. rates that are used; 10 Mbit/s which with a Hardware, software and staff expertise data payload of 8 bytes will result in a data costs would be significant and therefore throughput of 625000 bytes/s, 5 Mbit/s adoption of CAN FD for other automotive which with a data payload of 8 bytes will networking use cases is a possibility result in a data throughput of 312500 because it is only an evolution of the bytes/s and 2.5 Mbit/s with a data payload traditional and well-known CAN of 8 bytes results in a data throughput of technology. 156250 bytes/s.

Table 1 Comparison of key features of CAN, CAN FD and FlexRay

CAN-FD Bandwidth Potential In SEQ Figure 2 CAN FD is shown with an Arbitration Phase of 500Kbit/s and 1Mbit/s which sit below and above FlexRay at 2.5 A study was carried out to help understand Mbit/s. Therefore CAN FD is equivalent to more about the potential of CAN FD when low-end FlexRay. compared to FlexRay. As a reference traditional CAN is included assuming that it is running at 500Kbit/s due to the fact that the majority of passenger cars do not exceed this speed in their fastest CAN buses. In this study it was assumed that

100% bus utilisation could be achieved by using a pseudo-deterministic CAN bus scheduling technique [5] [6] or a deterministic CAN strategy such as Time-

Triggered CAN (TTCAN) [7].

02-21 iCC 2013 CAN in Automation

500Kbit/s Arbitration Phase, Max. Throughput vs Data Phase Bit Rate

CAN-FD Arb Ph. 250000 1Mbit/s

200000 FlexRay @2.5Mbit/s 150000 CAN-FD Arb Ph.

500Kbit/s 100000

(bytes per second) Maximum throughput 50000 CAN @500Kbit/s 0

500000

1000000 1500000 2000000 2500000 3000000 3500000 4000000 5000000 6000000 7000000 8000000 Bit Rate in Data Phase (bit/s)

Figure 2: Example complicated architecture that can be simplified using a high-speed networking technology

Figure 3: Summary of BWM X5 communications schedule

02-22 iCC 2013 CAN in Automation

Bit Rate (Bits/sec)

1G Ethernet

400M IDB-1394

TTP/C 25M MOST 25, 50, 150

10M FlexRay 3.0.1 / Byteflight CAN-FD 1M CAN /TTCAN

20K SENT Safe-by- LIN Wire

Relative Cost

Figure 4: Current and emerging communications technologies versus price versus bandwidth (based on Figure 1 of LIN specification package Revision 1.2 page 2 [8])

Real-Time Control Cycles The other interesting point to notice about this example of FlexRay communications FlexRay has a time-triggered is that the Dynamic Segment is used for communications schedule, each row of the flash download, diagnostics and communications schedule has a Static calibration and certainly one of the cited Segment and a Dynamic Segment. The motivations for CAN FD is for flash static segment is fixed and messages download in production and whilst vehicle designed to be transmitted there will is in service. As the Dynamic Segment is always be whilst in the dynamic segment 2ms out of a total 5ms communications there is the possibility to only send a cycle, then 60 % of the bandwidth cannot FlexRay message from an ECU if it has be used for flash download, diagnostics some new information to report. There are and calibration. always 64 rows to the FlexRay communications schedule. An example of Cost Comparison FlexRay communications schedule which summarises the use in the BMW X5 Previous studies have looked at how vehicle is shown in Figure 3SEQ. It can be features of microcontrollers impact upon seen that the communications cycle is the cost. In general things like memory 5ms long and messages are multiplexed capability and number of gates are within this schedule depending upon the aspects that can influence the cost of a transmission period required. It can be microcontroller. With this in mind the seen for this example that the required datasheets of a number of Robert Bosch cycle times vary from 2.5ms to 20ms. IP cores were studied and their features These cycle times would not be too are compared in in Table 2. From Table 2 challenging with CAN FD as a single it can be seen that FlexRay requires a message could be less than 100µs and at larger number of gates for its implemen- the same time transport a lot of data. tation and more message RAM in its ERay controller than the other cores.

02-23 iCC 2013 CAN in Automation

The C_CAN and D_CAN cores have lower The SENT protocol is a unidirectional point message RAM and number of gates. to point technology that is slightly faster CAN FD sits in between CAN and FlexRay than LIN (28 KHz) but also reported to be in its gate and RAM requirements. cheaper to implement. TTCAN has been assumed to be similar Gates, RAM Comparison – Bosch IP pricing to CAN since it uses the same silicon. FlexRay in its current revision Figure 3 shows a comparison between the v3.0.1, is expected to be significantly more relative nodal costs of protocol expensive than CAN, mainly due to a implementation versus its bandwidth for greater RAM requirement in the FlexRay the main automotive relevant protocols. controller. There is a spread in the cost This figure is based on Figure 1 of LIN domain for FlexRay due to the variety of specification package Revision 1.2 page 2, network architectures that could be used, but has been updated based on other for example bus and star configurations information already presented in this report are possible. Some configurations are to include the current and emerging lower cost, whilst others are of higher cost. protocols in the automotive industry in There is contradictory information in the terms of network technology costs. Since literature concerning FlexRay being a higher cost protocol [8]. In the year 2006, CAN is the de-facto automotive standard, the BMW X5 became the first production CAN is used as the relative cost of unity car to implement FlexRay within its and is therefore the reference protocol in chassis control system. Although the the figure. nodal costs of FlexRay are higher, an

interesting reason for choosing FlexRay is Table 2: CAN IP core feature comparison that it can replace several CAN buses, thus reducing wiring, connectors, Name Protocol Gates Message gateways and system partitioning effort RAM (Berwanger et al; 2005). TTP/C has been (Bytes) implemented on faster systems than ERay FlexRay 110000 8448 FlexRay and is slightly more expensive. It should be noted that Figure 4 does not C_CAN CAN 14500 544 capture the holistic costs and misses a lot of key information. It does not show data D_CAN CAN 18200 564 throughput for each network technology which does tend to be but does not M_CAN CAN/ 30700 4750 necessarily have to be directly related to CAN-FD bit rate. The received wisdom that LIN is M_TTCAN CAN/ 47700 5250 the low cost technology, CAN is the TTCAN medium cost technology and FlexRay is /CAN-FD an expensive technology is too much of a simplification. The general trend is that the higher the bandwidth, the higher the A LIN node is expected to be cheaper to cost of nodal implementation. In practice implement than CAN as a result of a many other factors must be considered number of cost saving features of the such as cost of training employees and protocol such as the use of a single wire buying tools, the architecture employed for the transmission of data (instead of the itself and warranty costs. twisted pair used for CAN), the ability that LIN slave devices can use cheaper RC Impact on Automotive Integrity oscillators (instead of crystal oscillators that are a necessity for all interoperable ISO26262 is the new standard providing CAN devices) and the physical layer is guidelines on how automotive electronic simpler and therefore cheaper in its systems should be delivered to provide an standard form. appropriate level of safety.

02-24 iCC 2013 CAN in Automation

Systems are rated by Automotive Safety Future Automotive Electronic Integrity Level (ASIL) from A to D with Architecture ASIL-D being the most safety critical. ASIL-D requires a Probabilistic Metric for Modern vehicles already have a large random Hardware Failures (PMHF) of number of different network technologies <10-8 h-1. It has been reported that within their electrical architectures such as traditional CAN does suffer from CAN, LIN, MOST and FlexRay. Ethernet undetected double bit errors with a certain is now being considered for some non- probability of this occurring [4]. This can critical high bandwidth applications and mean that under certain circumstances also in some safety critical applications. It traditional CAN does not satisfy the failure is extremely likely that due to the cost rate requirements due to the probability of sensitive nature of the automotive industry such undetected bit errors. This of course that there will always be a mixture of does depend upon the Bit Error Rate network technologies used in the vehicle (BER) of the transmission medium. It has and their sophistication will be due the also been reported that BER of an bandwidth requirements of the applications automotive CAN bus may change on the vehicle. Figure 5 shows a possible significantly over the life of the vehicle network technology mix and hierarchical such that failures become unacceptable architecture comprising of many for ISO 26262. established and emerging technologies FlexRay has been designed specifically appropriate to the different applications in with safety in mind and CAN FD has been the vehicle. designed with improved CRC handling and therefore it would be interesting to see how the two compare. Unfortunately this has been beyond the scope of this current study reported in this paper.

Figure 5: Future Automotive Network Technology Mix

02-25 iCC 2013 CAN in Automation

Conclusion Dr. Chris Quigley

Warwick Control Technologies Ltd. This paper has compared some of the key features and performance metrics Unit 8 Ladbroke Park associated with CAN, CAN-FD and Millers Road FlexRay. In particular data throughput, Warwick CV34 5AN cost of implementation and application to United Kingdom automotive electronic integrity systems [email protected] were compared. Finally a possible automotive network hierarchy of the future www.warwickcontrol.com was looked at. It is concluded that both CAN-FD and References FlexRay are likely to be a part of the [1] Hartwich F; “CAN with flexible data-rate”, automotive network technology mix of the th future. CAN-FD has the potential to Proceedings of the 13 iCC 2012 at Hambach Castle (Germany) provide the same data throughput as three [2] Berwanger J, Schedl A and Peller M to four traditional CAN buses. It also has (2005); “BMW – First Series Cars with the potential to provide a similar data FlexRay in 2006”; Automotive Electronics throughput to low-end FlexRay (e.g. Systems, Special Edition FlexRay, Page 6 FlexRay running at 2.5Mbit/s). FlexRay is to 8, Hanser Publishing. a faster network technology with some [3] Quigley, C., Jones, R. Peter, McMurran, R. safety features built in but its widespread and Faithfull, P. (2012) Microcontroller adoption is not yet guaranteed. Currently memory capability assessment and variant only a few automotive OEMs have selection for reducing the cost of adopted FlexRay and this could be in part mechatronics. International Journal of due to the risk of a new technology, higher Vehicle Design, Vol. 60 (No. 3/4). p. 248. cost of tool and personnel training. CAN- ISSN 0143-3369 FD has the potential to threaten FlexRay [4] Tran, E; “Multi-Bit Error Vulnerabilities in adoption for lower-end application and the Controller Area Network Protocol”, could be used to extend the life of older Carnegie Mellon University, Pittsburgh, PA, May 1999 CAN based platforms and architectures. [5] M. Grenier, L. Havet, N. Navet, "Pushing Interest in Ethernet and TT-Ethernet is the limits of CAN - Scheduling frames with growing for higher-speed applications and offsets provides a major performance therefore there is a threat to FlexRay from boost", Proc. of the 4th European different sides of the bandwidth spectrum Congress Embedded Real Time Software and could delay or even halt its adoption. (ERTS 2008), Toulouse, France, January However it is possible that FlexRay has 29 - February 1, 2008 carved out its niche enough to survive. [6] Tindell K and Burns A (1994); CAN-FD is likely to generically become “Guaranteeing message latencies on st known as CAN as it gets absorbed into a CAN”, Proceedings of the 1 International new part of the ISO11898 specification. CAN Conference. [7] Führer T, Müller B., Dieterle W., Hartwich F., Hugel R., Walther M., Robert Bosch GmbH; “TTCAN Time Triggered Communication on CAN”, Proceedings 7th International CAN Conference; 2000.

[8] LIN Specification Package, Revision 1.2, th 17 November 2000, The LIN Consortium (www.lin-subbus.org).

02-26