My first computer was a Cosmac Elf. My first “Desktop” was a $6,500 HeathKit H8. An Arduino today costs $3 and has more of nearly everything—except cost and size—and even my kids can program it. I became an embedded software developer without knowing it. When that H8 needed bigger floppy disks, a hard disk, or a network, you wrote the drivers yourself—in assembler if you were lucky and machine code if your were not.
Embedded software today is on the cusp of a revolution. The cost of hardware capable of running Linux continues to decline. Raspberry Pi (RPi) can be purchased for $25. A Beagle Bone Black (BBB) costs $45. An increasing number of designers are building products such as Cubi, GumStik, and Olinuxino and seeking to replicate the achievements of the RPi and BBB, which are modeled on the LEGO-like success of Arduino.
These are not “embedded Linux systems.” They are full-blown desktops—less peripherals—that are more powerful than what I owned less than a decade ago. This is a big deal. Hardware is inexpensive, and designs like the BBB and RPi are becoming easily modifiable commodities that can be completed quickly. On the other hand, software is expensive and slow. Time to market is critical. Target markets are increasingly small, with runs of a few thousand units for a specific product and purpose. Consumers are used to computers in everything. They expect computers and assume they will communicate with their smart phones, tablets, and laptops. Each year, consumers expect more.
There are not enough bare metal software developers to hope to meet the demand, and that will not improve. Worse, we can’t move from concept to product with custom software quickly enough to meet market demands. A gigabyte of RAM adds $5 to the cost of a product. The cost of an eight-week delay to value engineer software to work in a few megabytes of RAM instead, on a product that may only ship 5,000 units per year, could make the product unviable.
Products have to be inexpensive, high-quality, and fast. They have to be on the shelves yesterday and tomorrow they will be gone. The bare metal embedded model can’t deliver that, and there are only so many software developers out there with the skills needed to breathe life into completely new hardware.
That is where the joy in embedded development is for me—getting completely new hardware to load its first program. Once I get that first LED to blink everything is downhill from there. But increasingly, my work involves Linux systems integration for embedded systems: getting an embedded Linux system to boot faster, integrating MySQL, and recommending an embedded Linux distribution such as Ubuntu or Debian to a client. When I am lucky, I get to set up a GPIO or write a driver—but frequently these tasks are done by the OEM. Today’s embedded ARMs have everything, including the kitchen sink integrated (probably two).
Modern embedded products are being produced with client server architectures by developers writing in Ruby, PHP, Java, or Python using Apache web servers and MySQL databases and an assortment of web clients communicating over an alphabet soup of protocols to devices they know nothing about. Often, the application developers are working and testing on Linux or even Windows desktops. The time and skills needed to value engineer the software to accommodate small savings in hardware costs do not exist. When clients ask for an embedded software consultant, they are more likely after an embedded IT expert, rather than someone who writes device drives, or develops BSPs.
There will still be a need for those with the skills to write a TCP/IP stack that uses 256 bytes of RAM on an 8-bit processor, but that growing market will still be a shrinking portion of the even faster growing embedded device market.
The future of embedded technology is more of everything. We’ll require larger and more powerful systems, such as embedded devices running full Linux distributions like Ubuntu (even if they are in systems as simple as a pet treadmill) because it’s the easiest, most affordable solution with a fast time to market.