Au1000 Board Design

Work done in 2002

AU1500 chip with
             Alchemy logo

Alchemy Semiconductor were acquired by AMD.

AU1000 chip from AMD

The AU1000 on the JSE board.

Alchemy Semiconductor was a startup from DEC’s exiting the chip business. They had worked with the Alpha and the StrongARM, both speed demons. Cadence had invested in the startup, but shortly after shipping silicon, they were acquired by AMD, where they would be sold off again.... At 400 MHz they were well ahead of many embedded chips—certainly ARM—even with DEC’s StrongARM at 233 MHz.

The Ampro M2 and M3 boards shipped with a Linux port, however, only the base board had circuit diagrams. To manufacture a compatible board would need a boot loader or monitor. I had written a monitor some years back for the 68000 and ported it to the MIPS IDT3052 on the IDT385 evaluation board as well as the Transputer. It lacked TFTP booting over a network. A MIPS monitor called YAMON was open-source at the time. The Au1000 was certainly a neat device plus the self hosted tool chain was pretty useful. The Timesys port included the root file system which was hosted on a NFS mounted volume.

Our design targeted a six layer board and was to be our first to use a BGA package. We desperately wanted a decent target with networking capability and plenty of serial ports. When a company (who asked us to develop the Welch Allyn server replacement software), were also keen to have a target about a tenth of the Ampro price, we looked set to go. A design was entered in OrCAD schematic and partially laid out. When we wanted to progress to the first milestone, which was a tray of Au1000 devices in a minimum-order quantity, the partners were suddenly a bit hesitant. We were in no position to finance a six month design, debug, manufacture project with at least two board spins. About this time, JSE, a local South African company (near Stellenbosch, roughly 750km from Port Elizabeth), had completed an Au1000 network card. They were happy to supply hardware for a fifth the Ampro price. The risk in duplicating a board at these prices quickly terminated our Au1000 design efforts. We collaborated on the Linux port.

The design was based on application notes from Alchemy Semiconductor for their PB1000 evaluation platform. We also connected up the obvious; D0 to D0 for memory, address lines to the proper places. We used AMD Flash rather than Intel to try and reuse previous programming software from the ColdFire work. The Ethernet PHYs were Broadcom devices, similar to the Ampro M2 and M3.

Our attempt to design an Au1000 system was described in Au1000 Processor Board Design—au.pdf (1,45 Meg).