Cute. I’ve only skimmed the paper but if I use my 5 gate delay per instruction cycle rule of thumb I come up with a 20 IPS computer given a 10ms gate delay, or a 10 kIPS computer based on a 20us (50 kHz) gate delay. By comparison the KA-11 which was the CPU behind the PDP-11/20 circa 1969 was roughly a 2us cycle time (give or take, I’m being conservative here) or 500 kIPS. Based on this cite from page 8 of the paper:
Our demonstration is also slow, with each gate operation lasting at least 10 ms. Both speed and critical energy can be greatly improved through miniaturisation and material engineering. As shown in the Supplementary Information, reducing the width and thickness of the resonator to 2 µm and 10 nm, respectively, and decreasing the tensile stress by a factor of 100, would allow the Landauer limit to be reached and the speed of gate operations to be increased to around 50 kHz. Replacing the silicon nitride device layer with graphene would allow further miniaturisation into the nanoscale [45–47], increasing the gate speed into the gigahertz regime while maintaining Landauer limited operation.
I’m also assuming wire delays are factored into their gate delay estimates, but that may not be the case. I’ll hold off on the IPS estimate on a graphene nanomechanical computer until I see one. But I think this tech holds promise in high rad environments. High temperature environments remain to be seen, I’m somewhat skeptical there. But maybe finite state machines for simple electro-mechanical controllers. I think we’re a ways off from neural nets based on this technology. 🙂
The advantage I see with a nanomechanical computer is that if a Windows task hangs or a tab freezes, you should be able to just smack the side of the computer a few times and free things up.
When it comes to scrolling Windows, I’ve always thought a crank on the side of the monitor would be preferable to a mouse.
Cute. I’ve only skimmed the paper but if I use my 5 gate delay per instruction cycle rule of thumb I come up with a 20 IPS computer given a 10ms gate delay, or a 10 kIPS computer based on a 20us (50 kHz) gate delay. By comparison the KA-11 which was the CPU behind the PDP-11/20 circa 1969 was roughly a 2us cycle time (give or take, I’m being conservative here) or 500 kIPS. Based on this cite from page 8 of the paper:
Our demonstration is also slow, with each gate operation lasting at least 10 ms. Both speed and critical energy can be greatly improved through miniaturisation and material engineering. As shown in the Supplementary Information, reducing the width and thickness of the resonator to 2 µm and 10 nm, respectively, and decreasing the tensile stress by a factor of 100, would allow the Landauer limit to be reached and the speed of gate operations to be increased to around 50 kHz. Replacing the silicon nitride device layer with graphene would allow further miniaturisation into the nanoscale [45–47], increasing the gate speed into the gigahertz regime while maintaining Landauer limited operation.
I’m also assuming wire delays are factored into their gate delay estimates, but that may not be the case. I’ll hold off on the IPS estimate on a graphene nanomechanical computer until I see one. But I think this tech holds promise in high rad environments. High temperature environments remain to be seen, I’m somewhat skeptical there. But maybe finite state machines for simple electro-mechanical controllers. I think we’re a ways off from neural nets based on this technology. 🙂
The advantage I see with a nanomechanical computer is that if a Windows task hangs or a tab freezes, you should be able to just smack the side of the computer a few times and free things up.
When it comes to scrolling Windows, I’ve always thought a crank on the side of the monitor would be preferable to a mouse.