Open-Source Laptop and the Future of Embedded Systems
April 2, 2014, EE Live Conference, San Jose, CA—Andrew “Bunnie” Huang talked about the changes in the industry and the growing openness of consumer systems. Embedded systems are benefiting from the increasing commoditization of the underlying hardware.
In the earlier days of consumer electronics, in the 1930’s, products came with schematics, tuning specifications, and other documentation for the users. The Apple II included a reference manual that included schematics and source code. The average tinkerer could repair and modify the device without much difficulty.
Over time, however, the hardware became too complex and integrated for anyone to try to hack. The only function is to sit in front of your computer until something dies, then replace it. The continuing increases in integration are now turning in the other direction. The driver from Moore’s Law was to sit and wait for an improved device with greater innovation and performance. A T-flop is readily available in GPUs for games.
One major change is that the hardware is moving towards a common configuration as more of the systems are integrated into a CPU chip. In addition, changes in systems are implemented in days with software, while hardware changes take weeks to months. The software can be updated in seconds compared to the months to update hardware. The speed of change possible with software affects the adoption cycles, since new software can be available and in use in weeks to months while the hardware cycles remain in the year(s) range.
As a result, the hardware companies need to have 2 or 3 new products in the pipeline at all times to provide the improvements that the software demands. This situation eliminates all but the largest companies from producing processors. These companies have the access to the massive distribution needed to support the volume to justify the R&D for the next-generation processors. Secrecy then buys you a few months lead time before your competitor can reverse engineer the changes.
The issue is that increases the increases in performance due to Moore’s law are slowing down, so the computers are becoming more standardized. This standardization is easily visible in smart phones which are becoming static relative to design. As a result of the flattening of the change curves, now it is possible to use a linear approach to design change rather than the exponential one and still keep up with the desired improvements. If the Moore curve moves to 24 months, a linear approach is still even at 4 years, and a 36 month cycle permits 8 years for a technology cycle.
As product lifetimes stretch out to 6 years, the implication is that architectures become more relevant, and silicon and the development software and new code bases become worth porting. In the past, the hardware improvements gave a “free” 4-8 X improvement with no code changes. Now, the concerns for power efficiency at the same performance make new architectures essential.
The ability to make processors with custom features is leading to more custom ISAs and fine-grained optimizations are relevant. The ability to take 2-3 years to hand optimize code now makes sense, since the hardware cannot advance as fast as before. Specialty hardware is becoming more prevalent, as seen in ASICs that can mine Bitcoins.
Accompanying the changes in development emphasis from hardware to software are the considerations that style and fashion now matter much more. We are going from a focus on performance, size, battery life, and cost to one on style, fashion, “coolness”, usability, and the availability of relevant apps. The progressions of phones shows this trend. For years, Nokia made smaller cell phones. Then the iPhone appeared and the size has become a constant.
These drivers will put a greater value on craftsmanship and design. The increase in the repair and maker cultures is pushing everyone towards a fix-don’t replace mentality, because the newest gadget is not necessarily better or cheaper. In the post-Moore culture, there is less incentive to replace objects since many may have recycle value, and the ability to reverse engineer functions becomes greater.
Emerging markets are especially good candidates for the older technologies. The users are satisfied with technology that is 1-2 generations back, because it is sufficient for their needs. For example, many users in China are using phones with recycled parts from older phones and are built to older designs. The information ecosystem has enabled the recreation of feature phones. The software and detailed schematics are available for the older designs are easily accessed.
As a result, this area is becoming a fertile area for innovation. One phone can be plugged in to a computer and used as a mouse, another is for kids and has no keypad, so the kids can only call a few pre-selected numbers. The Chinese have coined a term for these small innovators, “shanzhai” which is like a hacker or maker. The shanzhai can work in small teams with little money to make new products. Although it took years to develop the ecosystem, the resulting stable and “open” systems provide documentation and development tools.
The emergence of open projects for hardware takes years to grow the infrastructure, but allows developers to start on shoe-string budgets and make products that are performance independent. The ecosystem still needs a good hardware supply chain, but is not shelf-life limited as in the past. The fact that wafer prices are increasing faster than the process nodes means that there is little benefit to moving to the next node after 28 nm. Therefore, the technology stack plus a small design team with time for growth, along with the growing do-it-yourself and repair cultures leads to an open hardware opportunity.
Developers are looking for simple, low-cost platforms like the Arduino, that is as powerful as many smart phones. FPGAs are approaching the performance of CPUs for computational tasks. Together, these are leading to a competitive do-it-yourself chassis for a notebook and tablet. Taken with the growth in the repair and restoration cultures foreshadows the emergence of “heirloom” hardware that is not designed for planned obsolescence. We should see hardware designed for beauty and lasting value and is upgradeable.
As a result, there is a new open hardware computer platform called Novena that integrates FPGA, standard peripheral ports, and other standard functions into a new platform. The documentation includes Gerber files and firmware and all are on-line at http://www.kosagi.com/w/index.php?title=Novena_Main_Page
The first PCB was designed for a server case, and the latest is geared towards a laptop configuration. Development of the new motherboard only took about 2 weeks, and the board is designed for mods and hacks. The modular design is designed to be upgraded when possible and individual components can be selected for any desired combination of performance and battery life.
The concept of an open computer design is possible because the components and the flattening of the costs makes a long design time feasible and realistic. Open hardware will allow designers to build on platforms as a labor of love, and not just to make next-generation boxes.
