Asynchronous processors

The processors used today, as is known from school and university computer science courses, are finite automata using Boolean (binary) logic. The problem is that it’s not possible to determine the state of such an automaton in its actual physical embodiment, since its cells have a nonzero switching time from one binary state to another (in other words, at the time of switching, the state of the cells and, accordingly, the automaton as a whole is undefined) . The clock signals are just needed in order to fix the discrete states of the automaton. In modern computers, the system clock generator (a separate specialized chip) is used for this, the signals from which are sent to all the chips installed on the system (motherboard) board. All the switches inside the microcircuits are attached to the leading or trailing edges of the clock signal, thanks to which the computer nodes operate synchronously.

Initially, all processors were synchronous. This went on for more than half a century, electronic engineers and programmers believed that it was natural and self-evident. Therefore, the clock frequency is one of the main characteristics of modern processors, although their performance is determined not only by it, but also by the architecture and instruction set. At a meeting of the Council of the Virtual Computer Museum, Yaroslav Afanasyevich Khetagurov, Chief Designer of Onboard Systems (BCVM) for the Navy, said that he was once summoned to the CPSU Central Committee and asked to explain why his BCVM performance was ten times lower than that of the American military using the similar element base. computer. It turned out that the Americans used very short teams in their car and “in parrots” it turned out to be much more productive, although in real combat missions both machines were approximately equal.

To begin with, we will define what a self-activated processor, self-activated logic, etc. are. The terms asynchronous processor and self-timed processor, clockless processor are synonyms, i.e. The same: a processor that has one or more internal clock generators and therefore does not require a central clock generator and circuits that support the propagation of a common external clock signal. Such processors emit much less radio waves and consume less energy compared to synchronous processors. When designing internal blocks of asynchronous processors, they are not considered to be synchronized with other blocks. For communication between the blocks, a single protocol is used: request - response with acknowledgment. When calculations are not performed, the blocks consume almost nothing while in sleep mode, but nonetheless respond instantly to requests. We see here that the calculation process is controlled by the data: no data - no computational activity. It is clear that such units are easier to integrate into so-called systems on a chip (SOC), and as a result, their overall performance is higher than that of synchronous processors. Increased productivity is achieved due to the fact that the unit does not need to wait for the arrival of an external clock signal to perform its work.

An asynchronous system is more reliable because its subsystems operate in wider ranges of variations in supply voltage and temperature. It is important that not only the processor be self-activated, but also the rest of the chips: the chipset, RAM, controllers, etc. The entire set of such chips is called self-activated logic (clockless logic). The American term clockless logic (literally - “unpolluted logic”) can be misleading, so we emphasize that for asynchronous microcircuits, clock signals are usually needed, but their source is local.

In addition to the already mentioned advantages of self-activated processors over traditional synchronous processors, there are a number of others. In the architecture, built on self-activated chips, the processor becomes quite simple and elegant. Many companies, including Intel, IBM and Motorola, conducted research in the field of asynchronous computing. The difficulties arising from the propagation of a common clock signal in ultra-large integrated circuits pushed for this. The first self-spinning processor was developed by Professor Elaine Martin (Alain Martin) at the California Institute of Technology (Caltech), and the idea itself belongs to one of the creators of computer graphics Ivan Sutherland (Ivan Sutherland), who wrote the first article on tactless logic. In 1990, a working group was set up in this area at the University of Manchester in England, and in 1994 she developed the first chip for cell phones. In 1997, Intel created a Pentium-compatible asynchronous test chip, which was three times more productive and consumed half the power. In 1998, Philips released an asynchronous processor for its pagers. In 2001, Intel in Pentium 4 partially implemented elements of asynchronous logic.

I note that every organ of the human body, every cell has so-called pacemakers, pacemakers that set clock frequencies, rhythms, according to which certain processes take place in them. Here we also see the asynchronous clocking scheme, to which the element base developers have arrived today.

Interestingly, if initially many ideas came to BT and programming from engineering and designing electronic devices, then there is a reverse movement: self-activated circuits are somewhat similar to OOP.

The ARM-Handshake joint product mentioned above is considered to be the first commercially available general-purpose asynchronous processor. Previous designs suffered from such disadvantages as low performance, programming difficulty and the problem of interaction with existing memory chips and peripheral buses designed for clocked processors. This product maintains compatibility with synchronous chips to solve this problem.

For the widespread introduction of asynchronous processors, no less than five years are needed, since they are still poorly supported by tools, and most importantly, they will require radical changes in the design of both hardware and software.