The First Fully Self-Timed SRAM Chip & Variable DC Power Supplier
The work on the fully self-timed SRAM has been presented at the last two PATMOS conferences (PATMOS2010 and PATMOS2011).
The conventional SRAM works based on timing assumptions, for example, at the first half clock cycle the pre-charge is done and at the second half clock cycle reading/writing is done. The timing assumption is set up according the statistic analysis. This means that there is a very high probability for data being read or written successfully. However for an individual reading or writing action, whether the above operations are successful is unknown, especially with DVS/DVFS for lower power and/or improving power efficiency.
Unlike the conventional SRAM, the fully self-timed SRAM is event driven based. All sub-operations, such as pre-charge, reading (WL), and writing-enable, are fully controllable. After one event finishes, the following events can be started, for example for data reading, only after bit-lines are pre-charged to a certain level, the following event, data reading, can start. This mechanism provides the following benefits: 1) shorten the width of the pulse for power saving purpose; 2) robustness to variation caused delay faults; 3) automatically adaptive to Vdd drooping; 4) working under variable Vdd down to 300mV; and 5) capable of being frozen at a certain Vdd level.
A 6T fully self-timed SRAM has been fabricated and tested. The chip is fabricated using the UMC 90nm CMOS technology. There are built-in testing circuits inside the chip and as well as an interface compatible with the conventional SRAM. There are a number of challenges for testing the chip to prove what we have claimed. And also there are many interesting phenomena we met during the testing.
To verify that the chip works under a variable power supply, a novel variable DC power supplier has been designed and fabricated. The power supply is called CBB, Capacitor Bank Block. By charging and discharging the CBB, a variable Vdd is provided. The power supply itself has other uses, for example to manage the energy obtained from harvesters and intelligently deliver the energy to tasks.
This demo will be arranged in the Wednesday afternoon at the poster session. All are welcome.