A little something to remember Mike Culbert

This morning I read on TUAW that “Mike Culbert, a longtime Apple hardware engineer, has passed away after battling cancer. […] His contributions include numerous patents for many iPhone and iPad innovations, we now take for granted. […] He was also a key player on the Newton development team.”

I want to pay my respects by unearthing a 1997 interview that originally appeared in Portable Design Magazine and was published at this link, which now appears to be dead. A few years ago I saved the article after it was briefly discussed on the NewtonTalk list.

Here’s the article, called PDAs: a study in contrasts.

From: Portable Design Magazine
Original link: http://pd.pennnet.com/Articles/Article_Display.cfm?Section=Archives&Subsection=Display&ARTICLE_ID=52439

PDAs: a study in contrasts

Portable Design interviews David Austin, director of engineering, Mike Culbert, systems architect, and Don Porter, mechanical engineer, at the Newton Systems Group (Cupertino, Calif.).

Alex Mendelsohn, Editor-in-Chief

* * *

PD: How did the MessagePad 2000 rise from the ashes of the original Newton?

Culbert: The idea was to design a PDA for the mobile business professional. The 2000 wasn’t envisioned to compete with an organizer, but rather to replace some of the common functions people do with a laptop – and do a better job at it. We looked at the forerunner Newton 110/121/ 130 Series and learned from its mistakes, taking a rototiller to the hardware and software and re-implementing the pieces which weren’t well done the first time around.

PD: How did you come to use the Cirrus Logic (Fremont, Calif.) Voyager chipset instead of the custom silicon used in the original Newton?

Culbert: We wanted a new chipset to expand the level of integration and eliminate jelly-bean parts.

PD: Did it let you make the unit physically smaller?

Culbert: No. But we knew users complained about the lack of PCMCIA slots on the original Newton. Without two slots, they couldn’t use storage and communications PC Cards at the same time.

We were already heading for a larger form factor to accommodate an additional slot. Our concern was how to reduce cost and power while boosting performance. Cirrus was looking for a partner to help it get into the PDA market. Since we needed a chipset, we decided to offer our PDA experience to Cirrus.

PD: Did Cirrus know you were going to use the Digital Semiconductor (Austin, Texas) StrongARM SA-110 processor?

Culbert: None of us knew we were going to use it when we started out. It didn’t exist. We were instrumental in its creation. We had this great architecture from Advanced RISC Machines (ARM – Los Gatos, Calif.), which was suited for Digital’s low-power process. Digital agreed, and we went from there.

That was totally independent of the Voyager chipset. The MessagePad 2000 design was already under way using an ARM710 microprocessor. However, once we had StrongARM silicon in hand, it became clear it was a viable alternative.

Very late in the StrongARM development game we did find a serious bug in the chip’s memory-management unit when handling permission violations. It took us and Digital’s engineers about a week to nail it down. What was astounding was that Digital delivered new silicon in less than four weeks.

PD: You were developing two microprocessor architectures for one end product?

Culbert: Yes, and it was difficult. We had to manage two logic designs in parallel to keep the MessagePad on schedule and to bring the StrongArm into the design without adding risk. At the engineering validation phase, one logic board ran a 25-MHz ARM710 and the other a 162-MHz StrongArm.

PD: That must’ve really complicated software development.

Culbert: It sure did. We had to support concurrent development and ended up building a lot of prototypes. It was an administrative headache to keep these platforms running in parallel and to make sure people had all their development needs met. But, our strategy was to stay on our time-to-market schedule and retain the opportunity to move to what we felt was the appropriate processor. If we couldn’t make it work with the StrongARM, we wouldn’t lose delivering the MessagePad on schedule. But, things did work. The Newton v2.1 operating-system software ran well.

PD: Were there technical aspirins to lessen the headaches?

Culbert: Sure. We were very careful to make all the code dynamic. For the most part, the software engineers on the team didn’t have to know what processor they were writing code for. Low-level software was implemented so that either one of the existing prototypes would function with it.

PD: Did you build in any special hooks to do that?

Culbert: Yes. We made sure the StrongARM and the Voyager included hardwired software-readable version registers. These were incremented on each version of the silicon to dynamically track low-level bug fixes and work-arounds.

For complex bugs it’s handy to observe what’s happening with a logic analyzer. We have a Hewlett-Packard (Loveland, Colo.) Model 16500B for state and timing analysis, but almost never use it during hardware development. I wrote an ARM disassembler for the H-P so we could look at traces of assembly code as it’s fetched. But I often felt that if I couldn’t debug my hardware with my trusty four-channel digital storage ’scope, it couldn’t be debugged!

We also used Geoport – an Apple protocol – for our internal debugging environment. It’s a high-speed protocol that uses SDLC framing at the low level but with a synchronous serial clock. It uses what would normally be hardware handshaking pins for the clock. Geoport let us use production-level system hardware and debug our ROM code using a Mac-hosted debugger.

PD: How did you hammer out the Voyager chipset specs?

Culbert: We spent about a year working with Cirrus on that. We created a common document and Cirrus implemented its chipset to that spec.

PD: Does the Voyager CL-PS7010 standard-cell ASIC handle SDLC and Geoport?

Culbert: Yes. It’s all in hardware.

PD: Why didn’t you use off-the-shelf SDLC/HDLC chips such as the industry-standard Zilog Z8530 serial communications controller?

Culbert: We used the 8530 SCC in the original Newton. But our goal now was to integrate as much as possible – for cost reasons. The chip we now have is bigger than we’d hoped for, but it offers significant cost savings over our previous architecture using separate ICs.

PD: Is Cirrus offering these chips to the merchant market?

Culbert: No. The Voyager is available only to Newton licensees.

PD: Does it make the MessagePad 2000 unique?

Austin: It gives us a high degree of differentiation by providing a lot of capability in a small space and with low power.

Culbert: The CL-PS7010 also has a proprietary low-pin-count bus and protocol that talks to the companion CL-PS7030 PCMCIA controller. We wanted to give users the ability to hot-insert and hot-remove PC Cards while the MessagPad 2000 was running. To do this we needed a completely isolated bus, so we came up with this scheme. Thanks to the cascadable CL-PS7030, the product can have as many as four Type II PCMCIA slots using one device per slot, all talking with a multiplexed frame format to the main controller. The bus offers voltage and signal isolation from the main system.

The unit’s DRAM and flash is the main system memory. It’s placed on the protected (memory) side of the bus, not on the PCMCIA side. Flash is where we store user information that’s not dependent on the unit’s batteries. We laid out the logic circuit board to use as little as 1 Mbyte of DRAM and 2 Mbytes of flash. Should there be customer demand, we may go to other configurations. We’re shipping 4 Mbytes of flash.

PD: How do you manage the flash array?

Culbert: There’s a flash file system, but not a Microsoft-type FFS. We use a transaction-based object database. It’s very effective for random access to large amounts of data.

Austin: If you put in a flash data card, you want the device to locate that as well as what’s stored in internal flash. The system automatically takes care of things stored in multiple areas – it provides a unified view. The transaction-based object store prevents data loss due to things such as a crash. Housekeeping is handled by the system controller.

PD: How did you engineer the analog interface to the unit’s loudspeaker, microphone, and touchpad?

Culbert: We spent months working with Cirrus-subsidiary Crystal Semiconductor (Austin, Texas) on the analog CL-PS7020 chip. We took a large discrete implementation and reduced board area, cost, and power by integrating it. Crystal brings a lot of patents in the area of sigma-delta modulation to the party. The ’7020 uses a linear single-bit A/D converter that operates in bit-serial fashion.

One of the problems inherent with sigma-delta is its slowness. It’s excellent for audio, but not very good for digitizing tablet signals where there’s electrical noise in the background and you want to detect and remove that noise from your measurement. Cirrus ended up filing four patents on the technique used in the ’7020. We got an A/D that’s speed-comparable to a successive approximation converter using a sample-and-hold, but on a very small piece of silicon – and with no external hold capacitor. It has a 19 micros sampling time.

PD: The unit has a unique subminiature connector. What is it and what does it do?

Culbert: It’s called the Newton interconnect. Originally we tried to fit an 8-pin DIN in there, but weren’t happy with the impact it had on our low-profile enclosure. Don Porter, our mechanical engineer, chose a 16-pin JAE Electronics (Irvine, Calif.) connector that was more suitable.

We added enough pins to it so a user would be able to use a communications device and a keyboard simultaneously, as well as providing charging power-in and peripheral power-out. That way, customers vertically integrating the unit could add meaningful backpack electronics. We ended up with 26 pins. JAE customized it for us.

PD: Did you work similarly with other suppliers for customized parts?

Culbert: We worked closely with Maxim Integrated Products (Sunnyvale, Calif.) to enhance its standard products. The Maxim variable-frequency and pulse-width switcher offers efficiency as high as 98%. It helps let the MessagePad work for up to six weeks from four AA cells.

PD: I can audibly hear the electroluminescent backlight switcher oscillating. Isn’t that irritating to users?

Porter: It’s one of our petty annoyances, too. It runs at a fairly low frequency between about 300 Hz and 1 kHz, and any mechanical resonance amplifies that noise. We tried freeing the toroids from the glue that holds them down, but we didn’t have enough room in the enclosure to make them float. We’ve had no customer complaints about the noise, but we’re working to solve it. At least it’s quieter than a hard disk.

PD: What makes the user interface unique?

Austin: You can rotate the screen for vertical or horizontal views. Horizontal’s good for a Web page, but vertical’s better for filling out a form. Software takes care of that. The software also gives two horizontal and two vertical presentations. That was done so extended PC Cards that might have antennas or dongles hanging out of them wouldn’t get in the way of the screen.

Porter: We also put key controls to the left and right so the MessagePad accommodates left- or right-handed users. We took the button bar that was silk-screened on the original Newton and put it in software. Users can now change what they want on it. That’s also good for licensees that want a customized user interface for vertical applications.

We also designed-in a full-size pen. Most PDAs have very thin styli or telescoping pens – people tend to play with them. We wanted a full-size stylus, but we had to fight for space for it.

We also put a real loudspeaker in, not a piezo speaker. A lot of work went into the 16-bit audio subsystem behind it. It actually runs GSM cellphone compression routines. Several compression engines permit high-quality voice at a low bit rate. That’s handled in realtime using only about 2% of the StrongARM’s processing resources.

Culbert: A big challenge was electrostatic discharge. The mixed-voltage design, where the lowest voltage is 1.6 V, didn’t leave much margin for ESD. So we routed a lot of external signals – including Reset – into general-purpose inputs on the controller to de-bounce them. We actually look for ESD events in software.

We also had to carefully control ground loops from ever seeing the ESD energy. A lot of PC Cards violate the PCMCIA’s specs. Card makers connect the metal structure of a card’s case to ground pins on the connector. That gives a chassis-to-digital-ground short with virtually zero impedance. As you can imagine, in this system, where the end of the card is exposed to the user, that presented a hazard.

We put a copper-foil shield on the PCMCIA slots under the connectors to provide a plane to absorb the energy coming in on a PC Card. We wanted to get the energy off the card before it got to the connector. We went through several iterations of the logic board design as well as the copper foil.

Inside the MessagePad 2000

The MessagePad 2000 uses a 162-MHz StrongARM SA-110 microprocessor coupled with a Cirrus Logic Voyager chipset. The SA-110 dissipates less than 500 mW, delivering 925 Mips/W. DRAM, used for temporary storage during program and system code execution, also provides the heap for a NewtonScript interpreted environment and a C heap, as well as stack and temporary buffer space for serial communications.

The CL-PS7010 packs systems-level glue functions, memory and single-/dual-scan LCD controllers, and serial ports for async and SDLC AppleTalk communications. It can support an external synchronous clock to run Geoport. The CS8130 is an IR transceiver. The CL-PS7020 provides a pen digitizer and touchpad interface, a battery and temperature monitor, audio record and playback, and a clock generator. The CL-PS7030 meets PCMCIA 2.01 specs, supporting hot insertion of 3.3-V and 5-V PC Cards. A 480 x 320-pixel indium-tin-oxide-coated glass touchscreen is used. The LTC1323 is a Linear Technology (Milpitas, Calif.) single-voltage RS-422 line driver.

Although the MessagePad was designed for 100% coverage boundary scan testing, 150 test points were requested by manufacturing. These were added late in the design cycle by carefully going around the Newton’s circuit board – by hand – turning vias into test points.

[Fig. 1] The interior of the Newton MessagePad 2000 reveals conductive-coated EMI/RFI shielding on the unit’s plastic case, relatively uncrowded circuit boards, dual PC Card controller ICs, and a mezzanine board for flash memory.

[Fig. 2] Apple’s Newton Systems Group engineering director David Austin (left), confers with systems architect Mike Culbert (standing, rear), hardware designer David Drummond (seated), and engineering manager David Turnbull.

Portable Design – July, 1997