An encore for Magic-1?
Shortly after I declared Magic-1 "hardware complete", I casually mentioned to
my wife that I was starting to think about Magic-2. Her response was
swift, and final:
"No, there will be no Magic-2!"
I can't blame her. She was an extraordinary good sport during Magic-1's
design and construction - especially during the wire-wrapping phase. For
most of a year, she put up with electronic junk littering the kitchen table,
wire-wrap insulation fragments on the floor and a husband often lost in
concentration while the kids were hollering for attention.
She's the love of my life, the woman I plan on growing old with, mother of my
children, my partner and best friend. I have to respect her wishes on
this.
So, there will be no Magic-2.
Instead, we'll call the follow-on project "Magic-16".
Magic-16
The Magic-1 project has been a great success - It's been a fantastic learning
experience. I started off with a set of goals and constraints, and the end
result has greatly exceeded my expectations. However, some of those
project goals and constraints required me to make trade-offs that I'd like to
address. In particular, two of my key constraints for Magic-1 were high
code density in my ISA design, and limiting the complexity to make it feasible
to build the project using wire-wrap and TTL.
The code density requirement led me to create an instruction set architecture
that was powerful and expressive - it packs lots of capability into a small
amount of code. I believe that for compiled C, my Magic architecture beats
x86 in code density by 5-10%. The cost of that, however, was loss of orthagonality
and elegance.
Much of the reason I wanted to do Magic-1 was to learn more about electronics
and hardware. However, the construction aspect of Magic-1 forced me to
avoid exploring some hardware tricks that would have been fun to try. For
example, pipelining instruction execution would be interesting, but was
impractical for Magic-1. Also, my sense of nostalgia led me to use TTL,
but FPGAs are very interesting to me.
So, before I move on to a totally new hobby space, I think I'd like to do
another CPU project to address some of these issues. Here is my first cut
at project goals and constraints:
 | FPGA. The system will be built using programmable logic.
Specifically, I'll be using an Opal-Kelly XEM3001 development board as
a starting point. Further, I will be expressing the design using Verilog
HDL and instantiating it with the Xilinx Web-Pack targeting the XEM3001's 400K gate
Spartan III FPGA. |
| RISC. The basic design will be a 16-bit Reduced Instruction
Set Computer. However, I intend this CPU to be fully functional and
capable of running a multi-tasking real-time operating system. This
implies a C compiler retargeting, user/supervisor modes, traps, interrupts and
memory management. |
| Whole system. There are lots of FPGA CPU designs around.
However, one thing that I haven't seen a lot of is a simple project that
covers the whole system from CPU through compilers, operating system,
middleware stack and application programs. One of the best parts of the
Magic-1 project was that it covered that entire space and really let me
experience first-hand the interplay between different parts of the system.
So, at the end of the day, I expect the Magic-16 project to yield a complete
stand-alone computer running a multi-tasking OS and at the very least serving
web pages and allowing interactive command shells. Oh, and of course
it's got to have lots of blinky lights. |
| Repeatable. I've had lots of email from people who would like
to do a project like Magic-1. Some have even expressed a desire to build
an exact copy of Magic-1. Though that would be possible, it would be
very difficult and potentially very expensive. I was lucky to get a lot
of Magic-1s parts cheaply as surplus, and the custom-made enclosure that
Alistair Roe made for Magic-1 would be horribly expensive to duplicate.
One of my goals for this project is that others could take the basic Magic-16
design and fairly cheaply and quickly replicate it. Not only replicate
it, but customize it - add new instructions, tweak the compiler, etc. |
| Short-time. I spent four years actively working on Magic-1.
I'm not interested in spending another four years on Magic-16. I see
this project as being completed in months rather than years. And, my
further goal is that once I have it done, someone else could follow along and
redo it over a weekend or two. |
| The starting specs. So, here's a rough recap of what I'm
shooting for:
 | CPU & most support functions in FPGA |
| Clean 16-bit pipelined RISC |
| Trap & interrupt support |
| User & supervisor mode support |
| LCC C compiler retargeting |
| Real-time OS (probably MicroC-OSII at first) |
| TCP/IP stack (probably uIP) |
| Native Ethernet interface |
| Small enclosure with lots of LEDs on front panel |
| $300-$400 budget |
| Easily customizable |
| Once complete, repeatable over a couple of weekends |
|
