| 摘要 |
A low-power timekeeping integrated circuit, using a double-buffered memory architecture: The user can freely read from user memory at any time, and an internal clock periodically updates a set of timekeeping registers. Transfer from the timekeeping registers to user memory (for update of the data) is performed as a block transfer, asynchronously and invisibly to the user. A special timing-window requirement is used to avoid access collision problems: each edge of the one-hertz oscillator signal is delayed slightly, and it is the delayed signal which actually clocks the update to the timekeeping registers. After a further small delay (long enough to allow for worst-case ripple-through delays in the timekeeping registers), a latched signal (NO-RIPPLE, in the presently preferred embodiment) is driven active. The signal NO-RIPPLE shows that any rippling has been completed and that access is safe. Thus, transfer will occur or not, but will never be cut short. When the user attempts to access the chip, the logic signal NO-RIPPLE is sampled. If NO-RIPPLE is not active, no transfers from timekeeping registers to the user memory is permitted, for as long as the user is accessing the chip. Whenever the user finishes an access, an update is automatically activated. Thus, in this scheme, the chip logic, while avoiding any updating during an access, will always provide the most current data into the user accessible registers, whenever possible.
|
