Hardware and Embedded Systems

Last Updated: Mon Feb 11 18:47:58 UTC 2019

How I designed the first Australian SPARC Workstation

Texas Instruments TI-58C Powered By USB

Product Design Notes, Brochures and Documentation

GCS/CTD Product Line Brochure - Front Page - Rear Page

Solbourne VKAB K-bus to VME-bus Adaptor Computer Technology Design / Graphics Computer Systems, 1992

Brochure for VKAB VME- Kbus Adaptor for S5/500 - Front Page - Rear Page Brochure for Solbourne S5/500 - Front Page - Rear Page - (VKAB and HDLC driver port) Solbourne S4000 - Front Page - Rear Page - Brief Page 1 - Brief Page 2 - (HDLC driver port only) Depicted here slotted into a Solbourne S5/500 chassis, this adaptor was designed to accommodate 6U VME boards in the "VME-less" narrow profile chassis. This product never made it beyond the prototype stage. The launch customer changed their mind about what they wanted. 100% basic hardware and mechanical design. The RHS image shows an installed 6U VME Ethernet adaptor, and the internal K-bus I/O board of the Solbourne which included serial ports, console port, monochrome frame buffer, and Ethernet AUI port. In a subsequent project I ported an SS1+ synchronous HDLC serial port driver (zs() device - Zilog 8530 SCC) to the I/O board, which also required 'stitching' additional clock lines on to serial ports (and glueing in extra RS-232C I/O chips). Some idiosyncrasies of the OS/MP kernel made this an interesting exercise. I later ported this driver to the S4000 motherboard running OS/MP 4.0. A related VME project which migrated to the Sun 4/260 was modification of the existing port of the CMC-1800 ISDN Primary Rate (30B+D) 6U VME card driver, which required the addition of diagnostics into the probe() code. Perhaps the most disappointing project I started during the GCS/Solbourne VME era was the intended port of the Sun 4/260 channel adaptor for the CRAY supercomputer, replacing the 4/260 with a Solbourne S5/600. After being told I had the project as lead engineer and would be doing the port and integration work, the customer balked at the last moment and bailed out. That was my nearest close encounter with the supercomputing hardware world.
MX Mongoose Type C Alpha 0, SPARC Motherboard Computer Technology Design / Graphics Computer Systems, 1992
Brochure for MX SPARC Workstation (Mongoose) Front Page - Inner Left Page - Inner Right Page - Rear Page

Two images here depict the first prototype of the Mongoose SPARC motherboard, a single M-bus, single S-bus slot IPC/IPX form factor workstation motherboard. This design is the first ever Unix workstation to be commercially manufactured in Australia, and the first ever SPARC workstation design Down Under.

I produced the prototypes in cca 6 months from starting, and took about 2-3 months to debug to an early production standard. Cables hanging off the CYM-6002K-40 CPU module belong to the HP-1651 Logic Analyser used during the debug. CPU and Mbus clock speed of this board was 40 MHz.

The most difficult problem we had to grapple with during the debugging was a sensitivity to clock skew in the motherboard chipset. This skew sensitivity was later found to be the result of chipset documentation errors which led to incorrect support circuitry designed around the the IC. Not our fault, but it caused us much pain before it was isolated and fixed.


The upper image is the Alpha 0 motherboard blank.
		The third Mongoose image depicts a prototype in late debug, fitted to a Mongoose specific fan and mounting tray in a CTD WorkPAC chassis, also a design of mine. Note the requirement for high airflow over the CPU dictates the use of a lateral array of fans, and interesting design problem in its own right. The PCB has eight layers and in its time it was the largest and most complex PCB in its class ever built in Australia.
Note that this example has the Sbus CG3 embedded frame buffer loaded, unlike the earlier prototype. The design is characteristic of the transitional period between pure surface mounted and pure conventional IC packaging. A mix of packages is used, with PALs, NVRAM, EPROMs all in conventional standard pitch sockets. Baseline project 100% hardware plus some of the SunOS drivers, and regression testing of SunOS port.


dmesg trace for Mongoose C VAC ENABLED in COPYBACK mode SunOS Release 4.1.3 (MONGOOSE) #2: Mon Dec 8 16:28:31 EST 2003 Copyright (c) 1983-1992, Sun Microsystems, Inc. cpu = GCS,MX10 mod0 = Ross,RT625 (mid = 8) mem = 65120K (0x3f98000) avail mem = 61022208 cpu0 at Mbus 0x8 0x220000 entering uniprocessor mode Ethernet address = 0:41:47:ee:0:1d espdma0 at SBus slot 0 0x400000 esp0 at SBus slot 0 0x800000 pri 5 (sbus level 3) sd0 at esp0 target 3 lun 0 sd0: <Quantum Atlas V QM309100XCLW cyl 9004 alt 2 hd 8 sec 246> sd2 at esp0 target 2 lun 0 sd2: <DCAS-34330 cyl 8203 alt 2 hd 6 sec 171> st0 at esp0 target 4 lun 0 st0: <Exabyte EXB-8505 8mm Helical Scan> ledma0 at SBus slot 0 0x400010 le0 at SBus slot 0 0xc00000 pri 7 (sbus level 4) cgthree0 at SBus slot 2 0x0 pri 9 (sbus level 5) zs0 at sys 0xf0200000 pri 12 (onboard) zs1 at sys 0xf0100000 pri 12 (onboard) SUNW,fdtwo0 at sys 0xf0900000 pri 11 (onboard) audio0 at sys 0xf0400000 pri 13 (onboard) MBus clock frequency = 40Mhz root on sd0a fstype 4.2 swap on sd0b fstype spec size 251904K dump on sd0b fstype spec size 251892K
		The Last Mongoose? Above and right: Mongoose C preproduction motherboard S/N 0009 (1994) remained in service primarily as an Xterminal (with 8-bit colour only unfortunately), running as intended in a Sun IPC chassis. The motherboard was loaded in Australia, the PCB made in Silicon Valley. This system was donated to me by GCS in 1996, and remained in service continuously until 2008, when it was no longer capable of properly supporting the bandwidth of more demanding X11 applications. It is loaded with a 90 MHz HyperSPARC, 64 MB of DRAM and a Quantum Atlas V internal hard disk, with a 4 GB BM external disk and an Exabyte EXB-8505 tape. Note the original GCS badges on the chassis. Over its extended operational life it has outlasted two internal SCSI hard drives, both of which failed due to the relatively high internal temperature in the Sun IPC chassis, not designed to be hard disk friendly. The chassis below the stack is a Trimm Industries 5.25" enclosure, the chassis below the monitor is a Digital Equipment Corp badged Trimm Chassis, with DEC specific ABS molded front panel.
The photographs were taken early March, 2006. As I have spare power supplies, spare IPC chassis and other components, this system was intended to be left operational until the motherboard failed, as an experiment to determine the longevity of this generation of hardware. While the main imperative in terminating the experiment was operational, I decided that having a live Mongoose in storage was better than having a dead one on display.

Sbus CG3 Colour Frame Buffer Computer Technology Design / Graphics Computer Systems, 1992
Brochure for MFB-ECL Monochrome Sbus Frame Buffer - Front Page - Rear Page - Schematic Brochure for CFB-A/GS 8-bit Colour Sbus Frame Buffer - Front Page - Rear Page
The Frame Buffer board depicted here was an enhanced clone of the standard CG3 board. In this instance I added a clever jumperable arrangement to select true greyscale video, or standard RGB video, using a network of mixing resistors. This meant that the board would provide true greyscale rather than "stunted" (green only) greyscale video when connected to the the standard SMI monochrome SS1 monitor, or third party greyscale monitors. Designing the resistor network to match impedances and provide the required mixing ratios proved most interesting. Another little feature on this board was a jumper to provide sync on green rather than SMI standard sync, again a feature which came in very handy in the OEM marketplace. As the board used the same LSI device as the vanilla CG3, this project was largely a hardware exercise, with minimal firmware hacks. The sibling Sbus BW2 clone board which I produced earlier required more substantial changes as I had to generate ECL video with Sun-3 compatible timing, replacing the Sun-4 video timing in the greyscale Sbus frame buffers.
This required a substantially new board design with new Forth PROM firmware, which I also had to produce. Without the funds to hire a logic analyser, the design was debugged with a 2 channel Tektronix scope. A high resolution variant of this board, the BW2HR with 1600x1280 resolution, never got past the paper design phase, the hi-res 1600x1280 CG6 derivative, and a CG8 derivative also never made it. Related products included a multiport ECL video driver box for testing monitors.
PizzaPAC SPARC Workstation Computer Technology Design / Graphics Computer Systems, 1992 Brochure for Scorpion Series Systems - Front Page - Inner Left Page - Inner Right Page - Rear Page

The 'PizzaPAC' marketing label was used for a range of 'pizzabox' desktop systems, using Twinhead or Integrix supplied motherboards and imported chassis. This example is an SS10 system using an Integrix chassis and motherboard, with a HyperSPARC CPU loaded. The Scorpion badge was created by graphic artist Mark Kopp (my since deceased youngest brother) in Perth.
TowerPAC SPARC Server Computer Technology Design / Graphics Computer Systems, 1992 Brochure for TowerPAC Peripheral/Server Chassis Front Page - Rear Page - (internal metalwork design only) The TowerPAC was an imported Taiwanese built Server case which GCS adapted by installing an CNC punched and folded internal tray designed to accommodate the SS10 or later SS20 motherboard. The tray also mounted a pair of low RPM 120 mm fans to blow airflow over the motherboard, while also including mounting holes for additional SCSI connectors. The TowerPAC's debut was comical. The MD, Geoff Croker, forgot to notify me that he intended to sell this product and sold the first unit while my engineering office remained blissfully unaware of the impending embarrassment. The CPU tray was designed, the prototype delivered, tested, installed and shipped to the client in three days. A very hectic three days, suffice to say. In hindsight the TowerPAC had the shortest gestation and development cycle of any product I had ever worked on. Note the small PCB on the SCSI connector, this was an adapter we designed which included SCSI terminators and a 0.1" pitch header for standard low density cable use. The depicted example has a dual CPU HyperSPARC loaded, and a GCS built CG3 Sbus frame buffer.

Scorpion / WorkPAC Server / Peripheral Enclosure Computer Technology Design / Graphics Computer Systems, 1992
Brochure for Scorpion System (Twinhead M/B) - Front Page - Rear Page - (WorkPAC chassis design) These two chassis were designed as OEM enclosures for SPARC workstations and associated peripherals. Probably the most important design feature in these products was the use of large low RPM fans to provide extra airflow for poorly air conditioned office environments, and low fan noise. Both were manufactured in Melbourne. The fishtank image on the monitor is the real thing, we built a fish tank into the carcass (case) of a dead RGB monitor.

DeskPAC-20 Server / Peripheral Enclosure Computer Technology Design / Graphics Computer Systems, 1992

Brochure for DeskPAC XX Peripheral/Server Chassis - Front Page - Rear Page Brochure for GoldPAC WGS Server Chassis - Front Page - Rear Page - (22 Carat Gold plated!) The DP-20 was used both as a peripheral chassis but also as a server chassis, with an installed CPU tray, unique to the motherboard used. DP-20s were used for the SS2, Mongoose C and SS-10 motherboards. Internal brackets permitted the installation of up to three 5.25" full height drives, two being installed at 90 degrees to the base of the chassis. Production quantities of the DP-20 ran into hundreds over several years. Variants included a stainless steel model, and a 22-carat gold plated model (which sold quite well - mostly to public service clientele!).
DeskPAC-5 5.25 inch Peripheral Enclosure Computer Technology Design / Graphics Computer Systems, 1992

Brochure for DeskPAC 5 Peripheral Chassis - Front Page - Rear Page Curiously, of all of my product designs, the DeskPAC-5 'shoebox' remained in production the longest, being built for at least seven years. Designed initially to reliably support the hot running Seagate Elite-9 5.25" drive in poorly airconditioned environments, the DP-5 proved to be very popular and was built in many variants. The final and most numerous variant, the 'Rev.EA' model, was optimised for minimal metalwork production cost, minimising the number of folds and punched holes in the metalwork, but also using improved internal airflow routing to maximise cooling effect. Designing good chassis can be challenging as it amounts to an 'origami game' in devising clever strategies to build up the box from as few parts as possible, with a minimal number of CNC operations per part.

32 Mbits/s High Speed Optical Datalink (HS-ODL)
Dataplex Pty Ltd, 1989

Following the completion of the DPX-610 my subsequent project was the later abandoned 32 Megabit/s rooftop optical datalink. At that time DPX manufactured a highly successful product line of low cost optical links, used to connect LANs between buildings up to 1 kilometre distant, at speeds up to 2 Mbits/s. The 'high speed ODL' was to extend the range and speed of the product line, and was an entirely new design using the telescope assemblies of the earlier product. This design reached the point of PCB layout but was shelved. Packaged on circular PCBs, the HS-ODL used a differential pair laser driver cascaded from a differential ECL buffer stage, and a variant of the low noise receiver used in the UWA optical link. The design included an infrared filter window using rare earth doped glass, with a multilayer bandpass interference filter, a measure necessary to cope with the background current arising from solar illumination - much the same design problem faced by designers of heat-seeking missiles. Unlike thermal dark current in fibre receiver detectors, background current can be a crippling source of shot noise in free space optical links. No material documenting this project has survived, unfortunately.

DPX-610 K-Mux Medium Speed Synchronous Mux
Dataplex Pty Ltd, 1989

Brochure for DPX-610 Intelligent Multiplexer - Front Page - Rear Page

The K-mux was my last and most complex communications equipment design, which absorbed much of my effort over an 18 month period. This design used no less than five Z-180 CPUs, and a pair of microcoded micromachines for generating framing structures. The K-mux was a fully software programmable intelligent synchronous mux which could independently manage its state and configuration data. It used cca 20,000 lines of embedded C and Assembler, which accounted for more than two thirds of the development time expended. The K-mux used an embedded serial packet oriented protocol scheme inspired by (read 'derived from') the DEC DDCMP protocol, via which the master CPU issued commands and interrogated the status of the up to four slave or datapump CPUs. One of the interesting features of this design was the level of tuning in the datapump code, where a single Z-180 at 8 MHz clock speed, using a pair of DMACs, was capable of pushing 72 kbit/sec full duplex and synchronous. Another interesting feature were the micromachines, which were loaded at boot or restart time with microcode which had been generated on-the-fly from decoded configuration tables in non-volatile RAM. Expansion I/O (datapump) boards in the K-mux included my design for a software programmable fractional frequency multiplier, which allowed the software to synthesize any frequency which was a ratio of two integers in the multiplier's operating range (this was done with a PLL). This provided the DPX-610 with the most flexible clocking facilities of any mux in its class at that time, globally. Sadly the K-mux never made it into production as local customers preferred less sophisticated and more expensive imports. Probably my most original from the ground up design, and certainly the most complex logic design.

DPX-610 Expansion PCB (up to three stacked)

Optically Isolated RS-422 Differential Serial Line Interface
Monash University Computer Centre, 1987

Between 1985 and 1987 I worked for the Monash University Computer Centre (now ITS) as a Computer Systems Engineer, responsible for maintaining VAX-750/780 systems, associated peripherals, and also all network installations for Clayton campus, including the design and project management of all of the early optical fibre links across the campus (Apollo token ring and Ethernet). My sole hardware design during this period was a line interface device, designed as a replacement for long haul current loop serial lines used to access some on campus sites. The single board device used a pair of opto-isolators, with the DTR pin powering the RS-232C receiver side, and a plugpack adaptor powering the line impedance matched RS-422 side. Over a dozen were hand built and used for specific links where network access was infeasible at the time, and current loop links too slow. The system ran up to 38.4 kbps but was usually operated at 19.2 kbps due to the propensity for XON/XOFF characters to be lost at the higher speed due to slow interrupt service rates. No photos are available, but I did have archived images of the VAX-780s it was used to connect - usually running DDCMP protocol over the link.

Type 84 140 Mbits/s 1.31 micron Optical Fibre Line Terminal and Repeater Equipment
NEC Corporation, 1984

NEC Emitter Coupled Logic + RF BJT / JFET

In mid 1984 I was hired by NEC Australia to participate in the design of the new single mode fibre 140 Mbits/s line terminal and repeater equipment then to be supplied to Telecom Australia (Telstra). The design work was performed during a 4 month period at the NEC Tamagawa plant in Tokyo, located in the area adjoining the Yokohama district. I was responsible for repackaging and adapting the existing Japanese design for the ITU spec export 140 Mbits/s equipment, specifically working on the optical receiver and transmitter boards. Including the line coding overhead, the baseband channel was clocked at 168 Mbit/s. These boards were implemented using NEC 600 series ECL logic devices which expanded my ECL design experience considerably - the NEC ECL was very similar to MECL III and delivered exceptional switching speeds. Unfortunately the end product remained unable to meet the 'defacto Milspec airborne' thermal environments specified by the customer, since the customer also insisted on convective cardcage cooling despite our repeated advice to the contrary. The receiver used an excellent NEC designed low dark current Ge or GaAsP avalanche diode, and NEC manufactured lasers with internal PIN diode power output sensors. NEC lodged several Japanese patents for circuit designs under my name at the time. Unfortunately I have no photographs of the equipment or the cards due to company security policy at the time. The depicted devices include 600 series ECL gates, bipolar single and paired RF transistors used in driver stages (common emitters in the paired device), and a JFET, all in high reliability NEC ceramic packaging.

Optical Television Transmission System (Analogue)
University of Western Australia, 1983

In the beginning ... My 1983 Honours project at the University of WA was my first attempt at a major system design, in this instance wholly analogue. I was supervised by A/Prof John Hullett, later the driving force behind the QPSX project. This project was also my apprenticeship as an Emitter Coupled Logic (ECL) designer, as ECL was required in the modulator and driver stage. The system was a Pulse Frequency Modulated design, using an MC-1658P Voltage Controlled Oscillator, an MC-10198P monostable and an MC-10210P as the differential pair driver for the MFOE-106F IR LED, although the design was built to accommodate a future laser device. The low noise wideband analogue optical receiver was derived from a UWA design used in OTDR equipment, and used a C30817 PIN diode and 2N4416A J-FET first stage. The system included FM stereo encoders and decoders, an audio amp, an FSK digital subchannel encoder/decoder and a range of other lesser features. The design was largely 'from the ground up' including the design of the linear power supply, system cooling and the cardcage/chassis assembly. The following photos were recovered by digitising archived 35 mm negatives - fortunately these survived almost a quarter century of storage, the hardware unfortunately did not.

Optical Modulator and Transmitter Module (ECL)

Low Noise Optical Receiver and Demodulator Module

Stereo Decoder Module

Subcarrier Module

Stereo Encoder Module

Audio Amplifier Module

Texas Instruments TI-58C Powered By USB
April, 2010

One of my favourite computational toys is the TI-58C/59 series of handheld calculators, which represented the state of the art around 1980. They remain amongst the ergonomically nicest calculators ever designed, the high contrast display being easy to read, and the keys giving the user confidence that one digit alone was entered. The principal obstacle to keeping the TI-58C/59 series alive is the fragile BP-1A battery pack, which is supposed to last up to ten years. The longest any BP-1A I have used has lasted is about 3 years. The last one I purchased was quite expensive. So I considered the problem of how to use cheap and easily available means to beat this problem for once and for all.

Initially I considered rebuilding the BP-1A to use three AAA sized rechargable batteries - elegant but time consuming and messy to implement. Ed Grochowski, a machine architect at Intel, devised a clean solution which exploits the fact that the TI-58C/59 series uses an 5V RMS AC power adaptor, internal rectifier, and the BP-1A as both a battery and defacto primary charge store capacitor for the power supply - his solution makes use of an external regulated 5VDC plugpack as a replacement for the existing TI AC plugpack, and discards the complete BP-1A module, plugging the hole with a plastic sheet.

Whilst I was impressed with Ed's solution, I lacked the patience to shop around for the required bits. This got me thinking - what is the most ubiquitous regulated 5 VDC supply available currently? The obvious answer is the USB which is these days used on almost everything. It took about ten minutes to rig a prototype cable, using the power connector off a TI plugpack, and a tatty USB 1.1 cable located at the bottom of my box of used cables. The installation works nicely, and the TI-58C runs much cooler as the BP-1A is not dissipating waste heat from the charging process. The final cable uses heatshrink tube to protect the +ve and -ve soldered cable splices, with additional black duct tape applied as a strain relief, just in case, and also protected by heatshrink. The BP-1A is gutted - easily done by using a finely serrated steak knife on the internal cross members retaining the three battery cells. The total cost of the installation is negligible, unless you choose to buy a new USB cable.

April, 2010

$Revision: 2.279 $