I am an engineer that works for Intrinsity. Our marketing folks asked me to speak directly to some of the points raised in this forum. Since the best technical coverage of what we've done in the latest edition of Microprocessor Report (which is a subscription service that not to many here probably get) here are some details to clarify thingst.
What we have developed over the past three years is the ability to design very fast dynamic (domino) circuits. This is done through a combination of our CAD tools, clocking, and a new logic family (the collection is called our Fast14(TM) Technology). Applying some constraints to the design problem allows us to automate the creation of dynamic (domino) layout much in the same way that others have been doing static for 15 years.
So, with all this innovation, we needed a proof-of-concept.. and this chip was our means.
The test chip has two processing elements, each with the following components:
64-bit 5-entry register file, providing 2 reads and 1 write per cycle.
2 32-bit ALUs providing single-cycle ADD, SUB, boolean, compare, and shift operations. This is a pretty complete ALU.
A 2KB data RAM
A 2KB instruction RAM
A sequencer.
The device is a small RISC core. It is programmed via scan, loading and unloading both the instruction and data RAMs. The device is controlled via external pins.
After fabrication, we had significant circuit yield at 2.2 GHz.
So, is 2.2 GHz really a great statement? Well, when you compare it to existing offerings at 1.8 GHz it doesn't seem particularly great. But when you get a little deeper you'll see that our frequency is rather impressive. Why?
we used a standard CMOS process.. no silicon-on-insulator, no copper, just a vanilla 0.18-micron 5-layer metal process.
the design took 6 months from spec to tapeout.
the ALU, register file, and sequencer designs required a only 3 logic designers.
Our tools and methodology provides fast circuits and a significant productivity gain.
With what we've learned, we're applying it to some products currently under development.
Oh yeah, point of interest. Most of this chip was constructed with CAD tools running on Linux. Our compute farm is over 150 strong and the standard engineering desktop is a Linux box.
As we're hard at work, any direct replies to this posting may be delayed. Thanks.
Slashdot Mirror
I am an engineer that works for Intrinsity. Our marketing folks asked me to speak directly to some of the points raised in this forum. Since the best technical coverage of what we've done in the latest edition of Microprocessor Report (which is a subscription service that not to many here probably get) here are some details to clarify thingst.
What we have developed over the past three years is the ability to design very fast dynamic (domino) circuits. This is done through a combination of our CAD tools, clocking, and a new logic family (the collection is called our Fast14(TM) Technology). Applying some constraints to the design problem allows us to automate the creation of dynamic (domino) layout much in the same way that others have been doing static for 15 years.
So, with all this innovation, we needed a proof-of-concept.. and this chip was our means.
The test chip has two processing elements, each with the following components:
The device is a small RISC core. It is programmed via scan, loading and unloading both the instruction and data RAMs. The device is controlled via external pins.
After fabrication, we had significant circuit yield at 2.2 GHz.
So, is 2.2 GHz really a great statement? Well, when you compare it to existing offerings at 1.8 GHz it doesn't seem particularly great. But when you get a little deeper you'll see that our frequency is rather impressive. Why?
Our tools and methodology provides fast circuits and a significant productivity gain.
With what we've learned, we're applying it to some products currently under development.
Oh yeah, point of interest. Most of this chip was constructed with CAD tools running on Linux. Our compute farm is over 150 strong and the standard engineering desktop is a Linux box.
As we're hard at work, any direct replies to this posting may be delayed. Thanks.