A Tale of Two Chips

Jeff Bier’s Embedded Vision Column

In 2012, I wrote about how mobile application processors were becoming increasingly popular in embedded systems. Since then, this trend has accelerated, fueled in part by low-cost development boards aimed at enabling embedded system developers to evaluate these chips and quickly create prototype products.
For embedded systems developers, these boards (some developed by chip suppliers and some from their partners) can seem like a dream come true: they’re inexpensive and energy-efficient, and they offer impressive features and performance. But working with these boards often puts me in the mind of Dickens‘ famous passage: „It was the best of times, it was the worst of times.“ On the one hand, these boards are easy to obtain and easy to get started with. (Where ‚getting started‘ is typically defined as compiling some code to run on the CPU.) On the other hand, developers trying to do more-sophisticated work with these boards can easily find themselves stymied. I see this most frequently in two areas: compute acceleration and I/O interfaces. One of the keys to mobile application processors‘ impressive performance/cost ratios and energy efficiency is the extensive use of heterogeneous processing: In addition to a multi-core CPU, mobile application processors include a bevy of more-specialized programmable and fixed-function processing blocks. These include GPUs that can be used as general-purpose parallel processors, video encode/decode engines, DSPs (often used for audio and sensor signal processing), and image signal processors (used to improve images captured by the image sensors). By virtue of their specialization, these co-processors deliver much better cost-performance and energy efficiency than the main CPU for certain sets of tasks. Therefore, utilizing the full capabilities of mobile processor chips (or anything approaching their full capabilities) requires harnessing these special-purpose engines. But doing so typically requires specialized knowledge, development tools and APIs. Unfortunately, mobile application processor suppliers often don’t make the necessary software and documentation generally available. Similar challenges abound when developers utilize sophisticated I/O interfaces and system control capabilities on mobile application processors. For example, in the past few weeks, one BDTI client has struggled with an incompatible USB3 port on one mobile SoC development board, and buggy power-management firmware on another. So, while building code for the CPU (which is usually a multi-core ARM) on a mobile SoC tends to be straightforward, building and integrating code that uses specialized co-processors and I/O interfaces is often an exercise in guesswork and frustration. This problem is exacerbated by the fact that while Linux is the default operating system of choice for many kinds of embedded systems, Android is the most important operating system in mobile devices. As a result, software development infrastructure for mobile processors is excellent for Android, but often spotty for Linux. Why aren’t mobile SoCs better supported for embedded systems? That’s a tale of two industries: the mobile chip business is driven by huge volumes, and is extremely concentrated (a few huge customers dominate demand). In contrast, the embedded systems business is driven by lower volumes (though with higher margins) spread across thousands of customers. Mobile chip suppliers are simply not staffed and organized to provide support for thousands of customers pursuing diverse applications. As growth opportunities shrink in mobile phones and tablets, however, some mobile processor suppliers are investing in embedded applications. Rather than try to quickly ramp up to engage with hundreds or thousands of customers, some suppliers are carefully choosing a modest number of embedded systems customers to support. That’s an understandable strategy – but to succeed with just a handful customers requires being able to pick the winners well ahead of the finish line – a difficult proposition. In contrast, suppliers with the fortitude to make the larger investments required to support large numbers of customers will be planting many seeds – and some of those seeds will grow into big opportunities.

Anzeige

Das könnte Sie auch interessieren

Auf der W3+ Fair 2018 vom 21. bis 22. Februar 2018 in Wetzlar dreht sich alles um Technologien, die künftig zentrale Rollen in den Bereichen Luftfahrt, Automotive und Medizintechnik spielen. Dazu treffen sich Experten aus Optik, Elektronik und Mechanik in der Rittal Arena zum Austausch und Networking: Insgesamt über 180 Unternehmen sowie ein breitgefächertes Vortragsprogramm erwartet die Besucher. Bis zum 8. Februar bekommen Sie kostenfreien Zutritt zur Messe. Dazu im Ticketshop den Code W3NL1601 eingeben.

www.w3-messe.de

Anzeige

Am 12. Januar  fand das 67. Heidelberger Bildverarbeitungsforum mit knapp 300 Teilnehmern statt. Die Sonderveranstaltung „10 Jahre Heidelberg Collaboratory for Image Processing (HCI)“ erfolgte in Kombination mit dem Physikalischen Kolloquium der Universität Heidelberg. Das HCI wurde 2008 als Teil des Zukunftskonzepts der Universität Heidelberg in der Exzellenzinitiative gegründet. Auf der Veranstaltung stellten junge Wissenschaftlicher des HCI Highlights aus den Forschungsergebnissen der letzten zehn Jahre vor. Abschluss der Veranstaltung war der Festvortrag von Prof.Dr. Carsten Rother als Nachfolger von Prof.Dr. Bernd Jähne. Seit November 2017 sind in der dritten Phase des HCI die Firmen Bosch, Bayer, Heidelberg Engineering und Volume Graphics beteiligt.

www.bv-forum.de

Das Unternehmen Jenoptik, Anbieter von optischen Hochleistungssystemen und Lösungen für Healthcare- und Industrieanwendungen, hat ein neues Applikationszentrum im kalifornischen Fremont eröffnet. Mit dem neuen Standort an der US-Westküste möchte das Unternehmen näher am Kunden sein und damit das Wachstum in den USA beschleunigen. Direkt vor Ort werden Jenoptik-Mitarbeiter zukünftig Unternehmen bei der Umsetzung von ihren Innovationszielen unterstützen.

www.jenoptik.de

The AdvaPix TPX3 has an array of clever pixel electronics capable of processing every detected photon. The camera is capable of measuring simultaneously position, energy and time-of-arrival. Rather than collecting data frame-by-frame the device generates a continuous stream of event data. The information about detected photon energy is used to create a full per-pixel spectra. The camera acquire up to 40 million hits per second.

www.advacam.com

Keynote-Speaker der 2. Industrial Vision Conference, die vom 21.-22. März in Ludwigsburg stattfindet, ist Jörg Heinze, Leiter Technologiefabrik Presswerk bei der Daimler AG in Bremen. Thema seines Vortrags ist ‚Qualität im Presswerk – Evolution in der Qualitätssicherung‘. Weitere Vorträge der Tagung stammen von Carl Zeiss AI, LMI, Matrix Vision, Micro Epsilon, MVTec, Stemmer Imaging, VMT … Empfänger unseres Newsletters erhalten den reduzierten Preis von 1.445€ statt 1.795€, wenn Sie bei der Anmeldung im Feld Bemerkungen das Stichwort „inVision“ eingeben.

www.sv-veranstaltungen.de

Das junge Unternehmen Nerian Vision Technologies verzeichnet weiteres Wachstum und konnte das Geschäftsjahr 2017 erfolgreich abschließen. Auftragsbestand und Umsatz forderten besonders im Bereich Entwicklung personellen Ausbau. Darum verstärkt Dr. Jan Hackenberg seit dem 1. Januar 2018 das Team bei der Optimierung der Softwareumgebung der Produkte sowie der Weiterentwicklung des Portfolios.

nerian.de

Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige