A flexible communication scheme for rationally-related clock frequencies

Abstract

Abstract-As a replacement for the fast-fading GloballySynchronous model, we have defined a flexible design style for SoCs, called GRLS, for Globally-Ratiochronous, LocallySynchronous, which does not rely on global synchronization and is based on using rationally-related clock frequencies derived from the same source. In this paper, using the special periodical properties of rationally-related systems, we build a latencyinsensitive, maximal-throughput, low-overhead communication method, based on the idea of using both clock edges to sample data at the Receiver. The validity of the method and its resistance to non-idealities such as jitter, misalignments and clock drifts are formally proven while experimental results including overhead are presented for 90 nm technology. Despite allowing much greater flexibility, the overhead of our method is comparable to that of state-of-the-art mesochronous communication techniques. We also show performances, complexity and overhead improvements over all other approaches that have so far been proposed for rationally-related clock frequencies.

Lowering the latency of interfaces for rationally-related frequencies

“…In a Globally-non-Synchronous system such as GRLS, the unknown skew among the clocks makes so that latency, intended as the time interval between data output at the Transmitter and data reception at the Receiver (with an ideal, zero-latency channel) is undeterministic: worst-case, average-case and bestcase latencies can be defined and calculated. We have already shown how the throughput and latency figures of our original GRLS interface are better compared to alternative approaches [14]. We now mathematically and experimentally show how it is possible to combine flow-control and synchronization considerations to come up with new synchronization interfaces improving up to 33% the average performances of the original GRLS interface without any complexity penalty.…”

“…The local clocks for the different islands are generated by the local Clock Generation Units (CGUs), which divide the frequency of the global clock by a programmable ratio. We have also presented a variation of GRLS which uses multiple global clocks running at rationally-related frequencies, increasing the flexibility of the system [14]. This solution is still equivalent to the original GRLS as all frequencies are rationally-related and are all submultiples of a frequency f H .…”

“…However, in this case the frequency f H does not necessarily correspond to the frequency of any physical clock. the original GRLS interfaces that we presented in [14]. In a Globally-non-Synchronous system such as GRLS, the unknown skew among the clocks makes so that latency, intended as the time interval between data output at the Transmitter and data reception at the Receiver (with an ideal, zero-latency channel) is undeterministic: worst-case, average-case and bestcase latencies can be defined and calculated.…”

“…This is a direct result from the fact that the problem of asynchronous communication is inherently difficult, because no assumption can be made on the clocks involved in communication [13]. We observed that by restricting the local frequencies to be rationally-related, it is possible to define a new design style, which we called GRLS for Globally-Ratiochronous, Locally-Synchronous, for which it is possible to design much more efficient interfaces, thus drastically reducing the area/power, performances and complexity overheads of traditional GALS methods [14]. We have also shown how the loss of flexibility compared to pure GALS, which translates to the inability to choose the lowest possible frequency for every block, has only a limited and reasonable impact on the power consumption of the system: by coupling GRLS to a quantized voltage distribution system and applying DDVFS [15], for Distributed Dynamic Voltage Frequency Scaling, it is possible to obtain energy benefits that are only ∼ 70% higher compared to those of an ideal, totallyunrealistic GALS design style, a sort of ideal boundary [16].…”

MTBF Estimation in Coherent Clock Domains

“…In the general case, no assumption is made on the values of ݂ ଵ and, ݂ ଶ , and every ratio is permitted according to the programmed values of the PLL multipliers. This case is similar to Globally-Ratiochronous, LocallySynchronous (GRLS) in [6]. The non-coherent scenario is illustrated in Figure 3, and corresponds to the case where the communicating clock domains are sourced from different references.…”