"One way race logic strives to save energy is by addressing the shortest-path problem. In one scenario, cars set off in multiple directions trying to find the fastest route. When the first car arrives, all the other cars stop, saving energy.Credit: B. Hayes and J. Wang/NIST" (Nist.gov/Brain-Inspired Computing Can Help Us Create Faster, More Energy-Efficient Devices — If We Win the Race)
Virtual neurons are terminals that store information for a while. The purpose of information storage or parking is to deny the situation that two or more data flows from different sources reach the same CPU at the same moment. If the CPU is busy another data flow or data package must wait until the CPU is ready, or the system can stuck.
There is one problem with the CPU. The problem is that the system can handle only one operation at a time. It is always possible that two data units are traveling through the system, reaching the CPU at the same moment. That thing causes a situation in which the CPU will get stuck. Solving this problem with single-processor systems is easy. When the system must detect an error. It simply sends the data in two routes. The other route is a little bit longer than the first route.
The CPU compares those data lines identically. And in that version the longer route guarantees that the data lines reach the CPU at different times. But in morphing neural networks, there is the possibility that the other processor sends data accidentally into the one processor at the same time. That can stuck locally in large and complex systems. And in that case, the system must remove data from stuck processors. But that destroys data mass.
In some systems, data travels through two data lines. Then those data lines send data through one CPU that compares those solutions. If solution or date flows are identical, there is no error.
There is the possibility that two data lines can reach the comparing processor at the same moment. If the system uses two data lines in the data process. To avoid that situation the system must store the information for a while. The CPU uses a terminal where the data can wait until the processor is ready for the next operation.
Another change is to make some data flows wait in terminals. The system loads data to the terminal. Data waits in those terminals until the CPU is ready for the next operation. When the CPU finishes operation. It sends information that it's ready.
That other version uses terminal processors or intelligent terminals. The intelligent router can transfer information first to the terminal processors. In that system, the router can send data to terminals using those processors one by one. When the terminal gets the information it sends information about that to the router, which knows that the terminal is busy.
When the CPU completes its mission it tells that to terminals, that they can send more data through it. The system can use TCP/IP to confirm. That data is not corrupted. In that process, the receiver sends information once back to the sender. Then sender confirms that the data is OK. And the sender can clear its memory. If there is an error, the system resends that data.
The terminals have numbers and when the router sends data to terminal number one, it marks that in its memory. Then that router sends another data package to terminal number two etc.. The terminal sends information to the router when it resends that data to the CPU.
https://www.nist.gov/blogs/taking-measure/brain-inspired-computing-can-help-us-create-faster-more-energy-efficient
https://en.wikipedia.org/wiki/Internet_protocol_suite
Comments
Post a Comment