可重構(gòu)處理器（Reconfigurable processors）已越來(lái)越多地出現(xiàn)在不斷快速更新變化的車(chē)載信息娛樂(lè)系統(tǒng)上。如今，這一技術(shù)正欲繼續(xù)拓展，進(jìn)軍高級(jí)駕駛輔助系統(tǒng)（ADAS）領(lǐng)域。而隨著自動(dòng)駕駛技術(shù)的出現(xiàn)，現(xiàn)場(chǎng)可編程門(mén)陣列（FPGA）及其他定制化處理器也將取得進(jìn)一步發(fā)展。畢竟，在人工智能發(fā)展變幻莫測(cè)的當(dāng)下，定制化服務(wù)已變得日益重要。
如今，越來(lái)越多的整車(chē)廠（OEM）開(kāi)始將軟件性能作為衡量車(chē)輛性能的標(biāo)桿。因此，那些能夠幫助汽車(chē)硬件作出及時(shí)調(diào)整，以適應(yīng)系統(tǒng)算法及車(chē)載軟件變化的設(shè)備，將變得更具實(shí)用性。對(duì)越來(lái)越多的系統(tǒng)來(lái)說(shuō)，軟件的升級(jí)可以帶來(lái)性能的大幅提升，而有時(shí)僅僅通過(guò)調(diào)整程序架構(gòu)，就能使硬件性能更上一層樓。
“算法一直在不斷推陳出新，然而研發(fā)人員發(fā)現(xiàn)，只有在硬件改變的前提下，汽車(chē)的性能才能提升2%，但是如果硬件不變的話(huà)，這就是空談。”鏗騰（Cadence）旗下Tensilica市場(chǎng)部的集團(tuán)高級(jí)總監(jiān)Steve Roddy談到，“研發(fā)人員需要一個(gè)可編程的解決方案，且該方案的計(jì)算力要足夠強(qiáng)大。”
在介紹2018款?yuàn)W迪A8的zFAS自動(dòng)駕駛技術(shù)時(shí)，奧迪就著重強(qiáng)調(diào)了可重構(gòu)硬件的重要作用，并再次明確了電子控制技術(shù)對(duì)程序計(jì)算有著非常高的要求，因?yàn)樵陉P(guān)鍵時(shí)刻，這將是攸關(guān)生死的問(wèn)題。英特爾（Intel）全新的Cyclone V FPGA將搭載ARM雙核處理器，可以有效提升奧迪控制單元的性能。其他一同搭載的重要部件還包括英偉達(dá)（Nvidia）的圖形處理器、Mobileye的視覺(jué)處理器以及中央處理器等。
“在這款?yuàn)W迪車(chē)上，我們首次將異質(zhì)性方式運(yùn)用到ADAS系統(tǒng)上。”英特爾汽車(chē)駕駛編程方案集團(tuán)（Intel’s Automotive Programable Solutions Group）高級(jí)總監(jiān)Michael Hendricks告訴我們，“ADAS和自動(dòng)駕駛的發(fā)展可謂突飛猛進(jìn)，幾乎每三個(gè)月就會(huì)有一項(xiàng)重大突破。專(zhuān)用集成電路（ASIC）的研發(fā)者很難迅速抓住應(yīng)用的要點(diǎn)，而FPGA則帶來(lái)了很多幫助，研發(fā)者也能更從容靈活地去適應(yīng)這些變化。”
這一技術(shù)巨大的發(fā)展?jié)摿?，正吸引著越?lái)越多的FPGA供應(yīng)商，Xilinx就是其中之一。借助可重構(gòu)設(shè)備，設(shè)計(jì)團(tuán)隊(duì)改動(dòng)處于研發(fā)階段后期的硬件時(shí)可以更省力。與此同時(shí)，設(shè)計(jì)人員也可借此設(shè)置控制單元，以適應(yīng)不同模型的需求。比如，有些ADAS系統(tǒng)會(huì)比同類(lèi)產(chǎn)品多幾個(gè)感應(yīng)器。
“有了FPGA，就能更好調(diào)整I/O（輸入/輸出），正確設(shè)置LVDS（低電壓拆分信號(hào)）和CAN連接。”Hendricks指出，“比如說(shuō)，F(xiàn)PGA可以提供海量的邏輯門(mén)，因此，可以對(duì)硬件與軟件進(jìn)行編程，從而優(yōu)化算法與硬件。”
定制化的硬件過(guò)去主要是用于研發(fā)的，而ASIC和專(zhuān)用處理器則投入量產(chǎn)。然而，在車(chē)載多媒體逐步取代無(wú)線收音設(shè)備后，定制化硬件的功能也隨之突破了自身局限。如今，隨著用戶(hù)對(duì)于娛樂(lè)資訊軟件的特性和功能提出了更多的要求，這些硬件能派上用場(chǎng)的地方也越來(lái)越多。在大型車(chē)輛內(nèi)，定制化的處理器甚至可用來(lái)支持人與人之間的溝通交流。
“在大型SUV中，汽車(chē)頂棚的智能聲音接收系統(tǒng)能夠識(shí)別誰(shuí)在說(shuō)話(huà)，并將其聲音通過(guò)揚(yáng)聲器播放出來(lái)。”Roddy說(shuō)道，“這些芯片可以實(shí)現(xiàn)很多此類(lèi)功能。”
目前，由于用戶(hù)對(duì)于汽車(chē)安全性和車(chē)載多媒體需求的不斷增加，汽車(chē)IC市場(chǎng)有著長(zhǎng)期穩(wěn)定的增長(zhǎng)預(yù)期。據(jù)IHS Markit預(yù)測(cè)，汽車(chē)半導(dǎo)體的市場(chǎng)總規(guī)模將會(huì)從2016年的321億美元上升到2017年的344億美元，較2015至2016年7%的增長(zhǎng)水平略有提升。
長(zhǎng)期的增長(zhǎng)預(yù)期，也在不斷推動(dòng)著汽車(chē)芯片市場(chǎng)的變革。英特爾收購(gòu)了生產(chǎn)FPGA的阿爾特拉（Altera），又收購(gòu)了制造視覺(jué)處理器的Mobileye。高通（Qualcomm）則在積極收購(gòu)恩智浦半導(dǎo)體（NXP Semiconductors），而恩智浦去年剛剛合并了飛思卡爾半導(dǎo)體（Freescale Semiconductors）。據(jù)花旗集團(tuán)（Citigroup）估算，英偉達(dá)（Nvidia）——這個(gè)原本并未大舉進(jìn)入汽車(chē)行業(yè)的公司，將會(huì)在2018會(huì)計(jì)年度在汽車(chē)這塊注資10億美元，想必也和2017會(huì)計(jì)年度該領(lǐng)域收益上升52%不無(wú)關(guān)聯(lián)。
盡管像英偉達(dá)和鏗騰（Cadence）這樣的業(yè)界新貴已開(kāi)始專(zhuān)注汽車(chē)應(yīng)用軟件，傳統(tǒng)供應(yīng)商依舊延續(xù)了其市場(chǎng)龍頭的地位。Semicast Research的數(shù)據(jù)顯示，2016年，恩智浦、英飛凌（Infineon）、瑞薩（Renesas）、意法半導(dǎo)體（ST Microelectronics）和德州儀器（Texas Instruments）依舊是全球最大的幾家自動(dòng)駕駛組件供應(yīng)商。而在位列前十的全球頂尖自動(dòng)駕駛技術(shù)公司中，還有博世（Robert Bosch）、On Semiconductor、微芯科技（Microchip Technology）、東芝（Toshiba）和羅姆半導(dǎo)體（Rohm Semiconductor）等。
Reconfigurable processors have seen growing use in rapidly-changing infotainment systems and are expanding into advanced driver assistance systems (ADAS). The role of field-programmable gate arrays (FPGAs) and other customizable processors is expected to develop further with the emergence of autonomy, where the vagaries of artificial intelligence makes customization more important.
As vehicle OEMs make software more of a differentiator, devices that make it simple to alter hardware to match changes in algorithms and software are becoming more practical. In a growing number of systems, changes in software can improve performance significantly. Sometimes, performance can be further improved by altering the programming structure.
“Algorithms change rapidly, researchers can find 2% better performance, but they can’t do that if they can’t change the hardware,” said Steve Roddy, Senior Group Director, Tensilica Marketing at Cadence. “They need a programmable solution that has the computing horsepower they need.”
The 2018 Audi A8’s zFAS piloted driving highlights the growing role of reconfigurable hardware, as well as the demanding computing requirements of electronic controls that make life or death driving decisions. An Intel Cyclone V FPGA that has dual ARM cores augments Audi’s control module, which also employs an Nvidia GPU, a Mobileye vision processor and a basic CPU.
“Audi is our first adoption in an ADAS system with a heterogenous approach,” said Michael Hendricks, Senior Director of Intel’s Automotive Programable Solutions Group. “ADAS and autonomy are moving so rapidly, there’s a major breakthrough every three months. It’s difficult for an ASIC developer to hit the bulls eye. FPGAs bring a lot of assets and give developers more flexibility to adapt.”
Expectations of strong growth is attracting other FPGA suppliers like Xilinx. Reconfigurable devices make it simpler for design teams to alter hardware late in development cycles while also letting them configure control modules that can be used in various models. For example, some ADAS systems may have more sensors than others.
“With an FPGA, you can tailor the I/O, setting the right number of LVDS and CAN connections, for example,” Hendricks noted. “FPGAs offer a sea of logic gates so both hardware and software can be programmed so algorithms and hardware are optimized.”
Customizable hardware was once used primarily for development, with ASICs or specialized processors used in production. But that limitation faded as radios transformed into infotainment systems. These devices are seeing more widespread usage in infotainment as users demand more features and functions. In large vehicles, customized processors will help foster communications between humans.
“In a big SUV, smart microphones in the headliner can figure out who’s talking and send their voice through the speakers,” Roddy said. “These chips let them layer in this level of functionality.”
Safety and infotainment are primary drivers behind expectations of continued solid growth for automotive ICs. IHS Markit expects the automotive semiconductor market from $32.1 billion in 2016 to $34.4 billion in 2017, slightly outpacing 7% growth from 2015 to 2016.
Long-term expectations are fueling major changes in the automotive chip world. Intel acquired Altera, which makes FPGAs. It also acquired Mobileye, maker of vision processors. Qualcomm is in the process of acquiring NXP Semiconductors, which last year purchased Freescale Semiconductors. Citigroup predicts that Nvidia, once a small player in vehicles, will hit $1 billion in automotive in fiscal 2018 after seeing revenue rise 52% in fiscal 2017.
Though newer companies like Nvidia and Cadence are focusing on automotive applications, traditional suppliers have maintained the bulk of the market. NXP, Infineon, Renesas, STMicroelectronics and Texas Instruments were the largest automotive suppliers in 2016, according to Semicast Research. Robert Bosch, On Semiconductor, Microchip Technology, Toshiba and Rohm Semiconductor rounded out the top 10.
Author: Terry Costlow
Source: SAE Automotive Engineering Magazine
