1.
Introduction
With the development of semiconductor technology, the size of the integrated circuit is reducing and the supply voltage is getting lower and lower, which results in increased soft errors caused by radiation. This leads to increased susceptibility of SRAM, latches and flip-flops to single event upset (SEU)[1, 2]. The energetic particle striking the circuit causes the transistor to generate electrons and holes. The generated holes and electrons are accumulated in a drain node, leading to an upset[3]. When the particle charge is beyond an expected range, a soft error will happen. The expected range is called critical charge. In other words, the higher the critical charge, the higher tolerant capability to soft error. The Semiconductor Industry Association Roadmap has identified SEU effects as the major threat to reliable operation of electronic systems in the future[4, 5]. Therefore, many schemes designed to tolerate errors have been presented, such as radiation-hardening-by-design (RHBD) techniques. RHBD techniques include spatial redundancy and temporal redundancy. Spatial redundancy techniques, such as triple modular redundancy (TMR) and dual interlocked storage cell (DICE) designs increase the reliability of the system at varying levels of performance penalties[5–9].
Fig. 1 gives the structure of a standard static latch, the latch is very sensitive to high energetic particles and easily causes an SEU. In the hold state, assume that the clock signal ck is 0, when an energetic particle strikes node A or node B, the logic state will upset and the fault value will be saved via the feedback until the next logic value is written.
onerror="this.onerror=null;this.src='http://www.jos.ac.cn/fileBDTXB/journal/article/jos/2018/5/PIC/17080018-1.jpg'" class="figure_img" id="Figure1"/>
Download
Larger image
PowerPoint slide
Figure1.
The standard static latch.
This paper proposes a new latch to mitigate SEUs, the latch includes an error detection circuit and a C-element. The error detection circuit produces signals when an error occurs in the latch. The produced signals were used to control the latch, thus the fault value will be blocked and the right value will be output. The C-element will correct the fault value in the error detection circuit if an SEU occurs.
2.
Previous work
C-element is a famous design which can block SEU induced soft errors, the basic structure of C-element consists of 2 PMOS and 2 NMOS. Fig. 2 gives the structure of C-element and its true table. When giving the two inputs the same value, the output will get an inverse value. If two inputs are of different value, the output will hold the current state. State holding characteristics can block SEU from occurring at the internal nodes with minimal hardware overhead[10].
onerror="this.onerror=null;this.src='http://www.jos.ac.cn/fileBDTXB/journal/article/jos/2018/5/PIC/17080018-2.jpg'" class="figure_img" id="Figure2"/>
Download
Larger image
PowerPoint slide
Figure2.
(a) C-element. (b) C-element truth table.
For mitigating the vulnerability to single node strikes, we propose a stacked transistor. The stacked transistor hardening technique has superior SEU tolerant ability[11–13]. The stacked transistor approach isolates the sensitive node connected to the two off-state transistors, if only one gets hit, the other one will prevent the critical node from flipping[14].
3.
Proposed scheme
Fig. 3 gives the structure of the proposed latch: the latch consists of two parts, the static latch circuit and the error detection circuit. In the static latch circuit, it has two transmission gates (T0, T1) and three inverters (INV1, INV2, INV3). Node A and Node B are connected to the corresponding location of the error detection circuit, respectively. There is a C-element and two stacked PMOS (M6, M8) in the error detection circuit. The error indication signal E0 and E1 are connected to the static latch circuit to control the state of the transmission path.
onerror="this.onerror=null;this.src='http://www.jos.ac.cn/fileBDTXB/journal/article/jos/2018/5/PIC/17080018-3.jpg'" class="figure_img" id="Figure3"/>
Download
Larger image
PowerPoint slide
Figure3.
The proposed SEU tolerant latch.
When the clock (CK) is high (the inverted signal CKB is low), the transmission gate T0 is turned on, T1, T2, T3 and T4 are in an off-state. We assume that the input D is logic 0, so node A is logic 0, node B is logic 1, node Q is logic 0 and node Qb is logic 1. In the error detection circuit, NMOS M9 is turned off and NMOS M10 is turned on. PMOS M5, M6, M7 and M8 are turned on due to CKB is low, NMOS M11 is turned off, node E0 and node E1 are connected to Vdd, so the value of E0 and E1 are logic 1. The output O is logic 0, is equal to node Q and the input D. When the clock (CK) is low (CKB is high), the transmission gate T0 is turned off, T1 is turned on, the latch is in a hold phase. In the error detection circuit, T2, T3 and T4 are turned on, PMOS M16 is turned on, PMOS M5, M6, M7 and M8 are in an off-state. NMOS M11 is turned on, because node B is logic 1, NMOS M10 is turned on, node E1 is connected to the ground and the value of E1 turns to be logic 0. E0 remains logic 1 because node A is logic 0 and NMOS M9 is in an off-state. Because the transmission gates T2, T3 and T4 are turned on, node P is logic 1 is equal to node E0, node J is logic 0. Node M is logic 0 is equal to node E1. After the signal passes through the C-element, node S and node N are logic 1. The signal will feed back node E0 and E1 via the feedback loop. Node E0 and node E1 are hardened.
It is assumed that node A has an SEU, node A holds a fault value 1 and node B holds a fault value 0 accordingly (node Qb is equal to node B is logic 0). So NMOS M10 is turned off and NMOS M9 is turned on. Node E0 is connected to the ground, and the value of E0 will be logic 0. Node E1 remains logic 0 because its charge has been put over before the SEU occurs. In the static latch circuit, NMOS M1 and NMOS M2 are turned off, PMOS M3 and PMOS M4 are turned on, the output O is equal to the node Qb is the right logic 0. Node B is similar to node A. It is assumed that an SEU occurs on node E0, node E0 holds a fault value 0, node J holds a fault value 1 accordingly, node M remains logic 0, as mentioned in section 2, the output of the C-element S will hold the current state. So, node S remains logic 1. The error will soon be corrected through the feedback loop. Node E1 is similar to node E0. In summary, the proposed latch can mitigate SEUs on node A, node B, node E0 and node E1.
When a particle strikes the error detection circuit. PMOS M5 and PMOS M7 may be turned on, the Vdd signal will influence the state of E0 or E1. The two staked PMOS M6 and M8 will prevent this case happening. If one PMOS was struck and turned on, the other one will block the signal. So, node E0 and node E1 are hardened again.
The clock uncertainty may impact this scheme due to extra switches being necessary for the scheme. However, the latch proposed in Ref. [21] includes 5 transmission gates controlled by the clock signal, it did not make any comment about the clock uncertainty’s impact to the scheme. Triple Modular Redundancy is the most common SEU hardened technology. The TMR latch in Ref. [22] has 6 transmission gates, the 6 transmission gates are controlled by the clock signal. There are more clock distributions in the latch, while the scheme can work well. It is the same in Ref. [5]. The proposed SEU tolerant latch in this paper has 5 transmission gates controlled by the clock signal, the number is similar to those in Refs. [21, 22, 5]. Generally, the clock uncertainty makes very little impact on the latch proposed in this paper.
4.
Simulation results
HSPICE was used to simulate the proposed latch, HSPICE uses advanced Predictive Technology Model (PTM). The nano-scale Integration and Modeling (NIMO) group of Arizona State University invented PTM. As an evolution of traditional Berkeley Predictive Technology Model (BPTM), PTM of bulk CMOS is successfully generated for 130 to 32 nm technology nodes[10]. PTM is more physical, scalable, and continuous over technology generations and suitable for emerging variability and reliability issues[15]. To simulate the circuits, a 32 nm technology model is used and the supply voltage is 0.9 V.
The SEU injection method is used to simulate high-energy particle striking. Fig. 4 shows the experimental results when an SEU was injected at nodes A, B, E0 and E1. It describes CK (the clock signal), the input signal D and signals on node A, B, E0 and E1. It is clearly in Fig. 4, the output O holds the same value of input D. Node A holds the same value of input D and node B holds the invert value. In the hold state, if node A or node E0 has an SEU, the output O is almost unaffected. If an SEU occurs on node B or node E1, the output O generates only a little glitch, which can be ignored. In general, the proposed latch has SEU tolerant capability on node A, node B, node E0 and node E1.
onerror="this.onerror=null;this.src='http://www.jos.ac.cn/fileBDTXB/journal/article/jos/2018/5/PIC/17080018-4.jpg'" class="figure_img" id="Figure4"/>
Download
Larger image
PowerPoint slide
Figure4.
(Color online) The experimental results when an SEU was injected at node A, node B, node E0, node E1.
The proposed latch is compared with other SEU tolerant latches in two terms: power and propagation delay. It is clearly in Fig. 5, although the latch proposed in the paper did not get the lowest power and propagation delay, it achieves lower power and propagation delay than the RHBD latch[10] and the TMR latch[17–20]. The proposed latch achieves lower propagation delay than the HPST latch proposed in Ref. [21]. Although the latch proposed in Ref. [16] introduces lower power and propagation delay, it cannot mitigate SEUs on node E0 and node E1, which will result in incorrect output.
onerror="this.onerror=null;this.src='http://www.jos.ac.cn/fileBDTXB/journal/article/jos/2018/5/PIC/17080018-5.jpg'" class="figure_img" id="Figure5"/>
Download
Larger image
PowerPoint slide
Figure5.
(Color online) (a) Power dissipation. (b) Propagation delay.
5.
Conclusion
This paper proposes a new latch that can tolerate SEUs. Through the error detection circuit, the latch can get indication signals and a fault on the error detection circuit can be corrected via a C-element. With comparison of other schemes in power and propagation delay, the proposed latch works well. Via the SEU injection experiment, it is obvious that the proposed latch is superior to other latches in SEU tolerant.
