Hardware specification

Hardware details for OLPC, March 14, 2006

Maintained from a document written by Michael Bove by Jim Gettys.

Hardware Design Process

Designing hardware is much more constrained than software; while you may sometimes have great influence on the design of a chip many months in advance of availablility, you can only actually use chips which you can get in the volumes required at prices that you can afford. Even a single missing component, or component not available in the quantities you need, may cripple your production. Many in the software community, who are used to more fluid ability to modify design and produce in unlimited copies, find this a foreign concept.

Designing hardware is similar to making sausage: you may be able to grow new ingredients starting long in advance if you are friendly with farmers (chip designers). You can only make your sausage, however, with the ingredients required by your recipie that you can actually buy in the volume you need to manufacture. Sometimes you can substitute ingredients without spoiling the general recipie, and sometimes the result would be inedible. In this case, we have a single chip that Mark Foster is specifying, that sits between the CPU and the display, and over which we have detailed control.

If you'd like some insight into this process, you can look at older versions of this page in the wiki.

High-Volume Design and Manufacturing

Furthermore, production of high-volume hardware is now a very specialized business, and is now often joint between the organization/company that specifies what the hardware should do—often to the point of selection of major and minor components—and an ODM (original device manufacturer), which specializes in very high-volume design and production. The ODM generally does the detailed design for production; e.g., exact part selection if there are variants, schematics, layout, board routing, mechanical design, testing, debugging for production, logistics, and production of the finished goods.

In OLPC's case, the ODM is Quanta, as announced in mid December. There is a good chance that your laptop was manufactured by Quanta, headed by Barry Lam, which is possibly the largest company few people have heard of. Quanta manufactures more laptops than any other company in the world (almost 1/3rd of the total made), whether branded HP or Apple or others. Detailed design of the first production OLPC design is just starting, though OLPC has investigated (and continues to investigate) the possible components and other design tradeoffs.

Note that CPU chip manufacturers generally provide sample designs, development boards, and application notes, that are often complete and usable by themselves, though often include interfaces or hardware you might not choose in volume production. These clarify how their products might be "designed in" to actual products. Our prototype machine seen at Tunis was using one of the AMD "Rumba" boards. It approximated much of the first OLPC hardware, though used a conventional disk rather than NAND flash, and has components we will not use (e.g. ethernet), and that conceptual (but working) model lacked the much cheaper flat panel that is under development.

Detailed schematics and layouts of such sample AMD designs are generally available in the chip manufacturer's developer programs. If you are interested in exact design details of hardware could you get for immediate experimentation, we direct your attention to these programs, which generally include the ability to buy such sample hardware. Most of the information required to program devices, however, is completely freely available at the manufacturer's web sites in fully public specifications.

In concert with ODMs, such sample designs are generally customized to fit the exact product needs and engineered for high-volume-production tooling and techniques that are not applicable to low-volume development-board runs. OLPC has just entered in partnership with Quanta on this engineering-for-production phase of the project .

Detailed schematics and board layouts of these high-volume designs are often considered proprietary to the ODM's, or jointly owned by both parties involved. They represent the competitive advantage one ODM may have with its rivals (who may have access to the same components as they do). Those design schematics are sometimes available to programmers after production starts under NDA agreements; for example, schematics of many of the iPAQ handhelds were made available to programmers in the open-source community under NDA, when insufficient written programming information was available. OLPC will try to document our designs sufficiently to avoid NDAs; we expect this will be less effort than the logistics of requiring NDAs in such a large and diverse community.

Forseeable Designs

Currently we can forsee three generations of machines: a first one to ship in late 2006 or very early 2007, a second production run sometime in 2007 that will incorporate a newer AMD chip and possibly a newer wireless chip, and an E-Ink (or other low-power, bistable display technology) based machine to ship when this new display technology is available at an appropriate price point. The further out, the fuzzier the crystal ball.

We will try to keep this specification up to date as more and more details of the first design (and subsequent designs) are nailed down, provide links to specifications for the chosen components, and provide information required to program them (e.g. address space assignments). The first generation design uses already available components, with the (major) exception of the new flat panel. The electrical interface to the flat panel is not yet complete.

Subsequent OLPC designs may use components that have not yet been shipped by their manufacturer, and we may have to arrange a program whereby the open source community can get early access to specifications of those components for driver development.

First Generation System

Though it has been criticized for having no hard drive, it does have a USB port, into which a flash drive can be inserted, having memory of 286,000 mb and retailing in the U.S. for $20.

Details of the first generation are firming up, and are closely related to the CPU and CPU support chip chosen:

• Processor: AMD Geode GX500@1.0W with AMD CS5536 companion chip (note that the "500" chip really operates at a 366MHz clock).
• Memory: 128MB DRAM, DDR-266
• Nonvolatile storage: 512MB (possibly 1GB) NAND SLC type flash memory
• Batteries: 8 AA NiMH batteries in a removable hard pack; we need to ensure good continuity between batteries and make swap of battery packs easy. We hope these hard packs can be field repaired in the case of battery failure. External charging circuitry. Other battery configurations may become possible in the aftermarket so long as they meet the physical and connector requirements. NiMH batteries can be charged with voltage spikes from human powered generators: LiION are critically concerned with charge voltage, and so are not suitable for such chargers.
• Power: 100-240VAC, 14VDC, also human powered chargers (e.g. crank and others) based on a permanent-magnet generators. After extensive analysis, building a generator into this unit now appears not advisable on ergonomic, mechanical and electrical robustness grounds. External generators opens up many more options for charging and also more battery possibilities.
• Audio: AC97 codec, probably the Analog Devices AD1888 contingent on final power testing, built-in stereo speakers and mono microphone, jacks for external stereo speakers and microphones, Line-out, Mic-in, all on the edge of the unit
• Graphics: Embedded in the GX2. It does include alpha-blending; does not support hardware 3D acceleration (e.g. no shaded triangle support or geometry pipe), graphics memory shared with main memory
• External ports: three external USB2.0. On-The-Go functionality is unlikely unless the errata on the CS5536 in this area can be engineered around. 500mA maximum total power supply (enough for one port drawing full power).
• Display: novel dual-mode, based on a 7" diagonal 4:3 aspect ratio TFT panel
  • Monochrome (high-res, low-power "E-Book") mode: 1110x830, reflective (ambient light, up to and including full sunlight)
  • Color mode: 640x480, quincunx-sampled, LED-backlit
• DCON (Display Controller) ASIC chip, to adapt the Geode to the flat panel. Details of its design are still being specified and will be available as its specification completes.
• Input devices: keyboard, trackpad (supporting pointing, scrolling, and possibly graphical input), 10 additional buttons adjacent to screen for use in E-Book/tablet mode, in two groups of 5. We hope that all buttons will be fully distinguisable (that chording of keys will be possible)
• Wireless: We have chosen the Marvell 8388 chip due to its ability to autonomously forward packets in the mesh even if we power off the CPU. The antenna design is considered one of the key on the OLPC machine: in most designs, they are afterthoughts. This may increase by a factor of 4 the area covered by a machine in the mesh over the typical commercial laptop.
• Adaptor, Inlet 2p 14V 12W, w/LED, wall mount type (like Nokia 770), conventional connector like that found on most laptops (such a charger might be able to be used unmodified if it matches), but much more careful physical design on the connector, since on some laptops this has been a failure point.
• Embedded controller: based on ENE 3920
  • Boot ROM: 8MbSPI serial Flash ROM, interfaced via embedded controller; we'd like to squeeze to a 4Mb part if this is feasible to save cost, but we don't yet know.
• BIOS/loader: LinuxBIOS is our intended BIOS for production units.
• Button/Switches: Lid close, Power
• Reset button near one of the bottom corners of the display assembly
• Status indicators, visible with lid closed:
  • Power
  • Battery status
  • Wifi active
• 10W maximum internal heat dissipation; 12W total (remember, USB is external): no fans or moving parts
• keyboard, 80 keys, 1.2mm stroke, rubber type, water resistant, dust proof. We are evaluating these keyboards to ensure they are satisfactory.
  • KB encoder: PS2 interface (since lower power than USB)
  • Indicators: Caps, Num, scroll locks.
  • Touchpad: Synaptics (to be confirmed)
• Weight target: < 1kg total
• Environmental Specifications
  • Temperature: somewhere in between typical laptop requirements and Mil spec; exact values have not been settled
  • Humidity: Similar attitude to temperature. When closed, the unit should seal well enough that children walking to and from school need not fear rainstorms or dust.
  • Maximum altitude: -15m to 3048m (14.7 to 10.1 psia) (operating), -15m to 12192m (14.7 to 4.4 psia) (non-operating
  • Shock 125g, 2ms, half-sine (operating) 200g, 2ms, half-sine (non-operating)
  • Random vibration: 0.75g zero-to-peak, 10Hz to 500Hz, 0.25 oct/min sweep rate (operating)

1.5g zero-to-peak, 10Hz to 500Hz, 0.5 oct/min sweep rate (nonoperating)

• Regulatory requirements - the usual US and EU EMI/EMC requirements, along with UL for safety
• Reliability requirements - we will put a significant number of units through a serious torture test, and make the system as rugged as we can afford. Exact specifications of these tests are TBD.

The remaining decisions should be made by the over the next few days and we'll try to keep this page up to date. Detailed design of the DCON chip will take a bit longer.

The display is the part of this design which is most novel. Exact details of its operation to come.

Second Generation Design

Second-generation unit will use a more power-efficient integrated Geode-based AMD chip (instead of the GX500/5536 set) and probably a next generation wireless chip. Feedback window on design of the next AMD chip will close in March 2006, so it's important to maintain dialogue with AMD regarding Gen 2.