This page contains information on a selection of my projects. Click
the thumbnails for full scale view. Much more information is available on
request.
|
|
|

(click for larger image) |
Developed 10Gb Ethernet architecture and simulation for 10G startup
and participated in series A fund raising technical due diligence.
This included overall architecture and modeling of:
- Use of the Obsidian simulink C/mex libraries for major
functional blocks.
- THP.
- AFE models.
- BLW
- Timing recovery.
- Startup controller and various state machines.
- Firmware architecture.
|
|
|

(click for larger
image) |
Developed C++ model of 1000Base-T architecture plus a simulink model.
- Filters, timing, magic packet BLW tolerance, split band filters.
- Developed fast simulink C/mex libraries of major functions.
1000BaseT Product Information
|
|
|

(click for larger
image) |
Developed detailed model of 100Base-TX architecture plus a simulink model.
- Filters, timing, magic packet BLW tolerance.
- Floating point model in simulink.
- Bit accurate models using fixed point models in simulink.
- Also did chip level spec for 100Base-TX chip for a customer.
Obsidian Product Overview
|
|
|
|

(click for larger
image) |
While at SMC I proposed a scheme for transmitting 100Mb Ethernet over
Category 3 twisted pair cable. This scheme transmitted data with a 3
level code on 3 wire pairs while listening for a collision on a forth
pair. This arrangement eventually became the ill-fated 100BaseT4
standard. I received a patent for the technology.
Text of article.
|
|
|
 |
Working as a contractor I defined the AFE for a 100Base-T2 became
familiar with the DSP architecture.
Also did bench debug of T2 chipset. |
|
|

(Click for full image) |
Per the library (below) I built some custom multipliers based on an
optimized cell library.
These were implemented using Tanner EDA tools and irsim for
verification.
|
|
|

(click for larger
image)
|
Developed a high efficiency library of schematic / layout cells for fast
implemented of custom DSP macro cells. Shown left is simple LMS filter
tap implemented using the library.
- 4x denser than standard cell implementation.
- 2x faster, 30% the power dissipation of standard cell.
- Use of ratiometric and half static logic.
- Easy to port λ based layout rule set.
|
Precision Band Gap
|
|
|
In 65nm CMOS this bandgap has a precision of ±0.1%.
Responsible for the design, simulation and revue of layout.
|
|
|
|
|
Negative Substrate
Regulator
|
|
|
65nm leakage reduction power supply. A charge pump and linear regulator
provide -0.6V to 0.3V at 8mA.
Responsible for the design, simulation and revue of layout on this
project.
|
|
Special IO Design/Layout
|
|

(click for larger
image) |
Responsible for a number of IO pad designs, including Crystal
Oscillator, PECL, 5V tolerant, LVTTL etc.
Responsible for the design, simulation and layout on this
on many projects.
|
|
|
|
|
8 Bit
125Ms/S Pipelined ADC
|
|

(click for full image) |
In 180nm CMOS this project was for an Ethernet analog front end.
Made from a stack of 1.5 bit switched cap ADC's. The most critical
element was the differential stage amplifier shown
here.
Excellent dynamic characteristics are needed for the system to settle
within one clock cycle. A folded structure was used to achieve this
within the 1.65V minimum VDD.
|
|
|

(click for larger
image) |
Developed innovative multiprocessor concept, produced software modeling
to test efficiency.
- Explored the architecture with a number of compute intensive
algorithms.
- Setup GNU based compile framework.
- Attempted funding through DARPA, NSA etc.
Effort concluded in 2001. However, the performance benefit is so
significant that this type of component is bound to evolve eventually.
Unfortunately, the hierarchical arbitration scheme was rediscovered and
patented by Intel.
rRam concept.
|
|
|
 |
Proposed an innovative re-configurable array processor to DARPA in
conjunction with UCI. Received $1.2M in funding. Personally, I:
- Developed the architecture of the re-configurable array
processing element.
- Developed the programming support software for the array,
including a text base assembler, and a GUI based assembly code debug
tool.
- Problems with this architecture led me to develop the
rRam concept.
A company (Morpho Technologies)
was set up to exploit this technology.
|
|
|

(click for larger
image) |
This is a system project to adaptively remove sound noise. My
contribution to this was:
- All of the DSP software in C and assembler.
- DSP algorithms for a to channel system.
- Analog board schematic design.
- Effective? Take a listen to the system turning on in the
presence of white noise:
MP3 demo.
- Software
project.
Product Page
|
|
|
 |
Shown at the beginning of top level routing this is a mixed signal chip
to interface Cell phones with car stereos through a standard USB cable.
Analog cells included 8 bit ADC, amplifiers, comparators, references,
special IO's etc.
Not a very complex ship, but all design work done myself.
|
|
|
|

(Click for full image) |
During the period 1983-1985 I ran the VLSI Design group at Acorn, later
to become Arm. My contributions included:
- Participating in the architecture development.
- Design of the ALU, register file, clocking scheme etc.
- Taking the initiative to sale the design rights and writing the
first licensing contract with VLSI Technology.
More details
of my participation.
|
|
CMOS Phase Lock Loop
|
|

(Click for full image)

(Click for full image) |
A basic CMOS Phase Locked Loop (PLL). One of several I have worked
on. This one I own and developed on Tanner.
- I developed the complete circuit using Tanner Tools.
- Includes circuit to insure solid startup, and prevent common
reliability issues.
- Built in reference circuit.
- Has the usual lock detect etc.
- See the data sheet.
|
|
Multi-Value-Logic Test Chip
|
|

(Click for full image) |
A test chip for a JPL/DARPA project to demonstrate features of a "simugate"
4 level logic scheme vs standard binary CMOS logic.
The chip consisted of 4 ALUs, each of 8 bits: 2 implemented in four
level logic, 2 implented in binary logic.
The custom 4 level logic gates and IO drivers were implemented to
tape out within four weeks. |
|
|
|
|
|