Cross-Entropy Optimization Method

Một phần của tài liệu Machine learning and intelligent communications part i 2017 (Trang 530 - 575)

The main idea of CEO for antenna selection is to iteratively update the proba- bility vectorp, which can be defined as

p=

p(1,1), . . . , p(1,M), . . . , p(k,j), . . . , p(K,1), . . . , p(K,M)

. (4)

wherep(k,j) represents the probability of thej-th antenna of thek-th Rx to be selected. To effectively update the probability vector p, there are two iterative phrases which should be carefully designed as follows.

1. The random mechanism to generate a sample of random data. In this paper, Bernoulli probability mass functionsf(Ωq;p) is used to generateQsamples

f(Ωq;p) = K k=1

M i=1

p(k,i)

Ωq(k,i)

, q= 1, . . . , Q. (5) 2. The way to update the parameters of the random mechanism. The probability

vectorpis updated according to the following equation piter =

Q

q=1I{S(Ωiterq )>=γiterqiter Q

q=1I{S(Ωiterq )>=γiter} . (6)

whereI{x} is an indicator function. When the conditionxis satisfied,I{x}= 1, I{x} = 0, otherwise. Ωqiter is the selected antenna vector for sample q at the iter-th iteration.S

Ωiterq

represents the minimum stream SINR of sampleqat theiter-th iteration.γiter=S((1−η)Q)is the (1−η)-th quantile in the sequence S(1) >=S(2) >= ã ã ã >=S(Q), ã is the ceiling operation, and η (0,1). To smooth out the values ofpand prevent some componentp(k,j) ofpfrom being zero or one in first few iteration, the smooth parameterλ∈(0.7,1] is introduced as follows

piter=λpiter+ (1−λ)piter−1. (7) Hence, the implementation of the AS technique based on the CEO algorithm in the TM scheme can be summarized as follows

Algorithm 1. Antenna selection based on the CEO algorithm

1: Determine the set of the optional antenna combinationsΦaccording to the feasibil- ity conditions of IA in the IA subnetwork, i.e., the number of the selected antenna NIAat the Rx side of each IA pair.

2: Setiter= 0 and initialize the probability vectorp0= NNIA1.

3: repeat

4: GenerateQsamples according to the random mechanism Bernoulli probability mass function (5).

5: For each sample, calculate the precoding and decoding matrices based on the MinIL algorithm, and the minimum stream SINR.

6: Sort the minimum SINRs in descending order.

7: Update the probability vectorpiterby (4), and smoothen it by (7).

8: iter=iter+ 1.

9: untilThe stopping criterion is satisfied.

4 Simulation Results

In this section, the asymmetric interference network with the configuration (M, N, K, d) = (3,3,5,1) is considered. Assuming that the path-loss exponentα is set to 3. All the pairs are randomly and uniformly scattered in a 1 km×1 km square area, and the distance between the transmitter and its corresponding receiver is set to 100 m.

0 2 4 6 8 10

0 1 2 3 4 5 6 7 8 9 10

Tx1 Rx1

Tx2 Rx2 Tx3

Rx3 Tx4 Rx4

Tx5 Rx5

Location in X (100m)

Location in Y (100m)

Tx Rx

IA Cluster SM Cluster

Fig. 2. The network topology after performing the clustering management scheme when the transmitted SNR is 12 dB.

A network shown in Fig.2is taken as an example to analyze the performance of the CEO-based AS method in the clustering management scheme. After per- forming the clustering management scheme in the network when the transmitted SNR is 12 dB, the whole network is divided into one IA cluster and one SM clus- ter. From the figure, we can observe that the four pairs, i.e., the 1-st, 2-nd, 3-rd and 4-th pair, jointly comprise the IA cluster, and the 5-th pair acts as the SM cluster independently.

According to the result of clustering, the spectrum efficiency of different schemes under various transmit SNRs is compared in Fig.3. From the simulation results, we can find that the CEO-based AS method can effectively improve the performance of the IA cluster compared to the original clustering management scheme. However, compared to the ES-based AS method, there exist a little performance gap. Considering the high computational complexity of the ES- based AS method, the CEO-based AS method can achieve a balance between the spectrum efficiency and the computational complexity.

−100 −5 0 5 10 15 20 25 30

5 10 15 20 25 30 35 40 45 50

Transmit SNR (dB)

Spectrum Efficiency (bits/s/hz)

MinIL

MinIL−based CM MinIL−based CM with CEO

Fig. 3.The spectrum efficiency versus the transmit SNR under different schemes.

5 Conclusions

In this paper, the implementation of antenna selection in the clustering man- agement scheme was analyzed. The redundant antenna resource was used to perform the AS technique. However, the ES-based AS method was unfriendly for performing in practical wireless networks due to the high computational complexity. The another effective combinatorial optimization method, i.e., cross- entropy optimization (CEO) algorithm, was selected as the substituted method.

Finally, the simulation results were presented to show the effectiveness of the proposed CEO-based AS method in the clustering management scheme.

References

1. Cadambe, V.R., Jafar, S.A.: Interference alignment and degrees of freedom of the K-user interference channel. IEEE Trans. Inf. Theory54, 3425–3441 (2008) 2. Gomadam, K., Cadambe, V.R., Jafar, S.A.: A distributed numerical approach to

interference alignment and applications to wireless interference networks. IEEE Trans. Inf. Theory57, 3309–3322 (2011)

3. Zhao, N., Yu, F.R., Jin, M., Yan, Q., Leung, V.C.M.: Interference alignment and its applications: a survey, research issues and challenges. IEEE Commun. Surv.

Tutor.18, 1779–1803 (2016)

4. Zhao, N., Yu, F.R., Li, M., Yan, Q., Leung, V.C.M.: Physical layer security issues in interference-alignment-based wireless networks. IEEE Commun. Mag.54, 162–168 (2016)

5. Zhao, N., Yu, F.R., Sun, H., Li, M.: Adaptive power allocation schemes for spectrum sharing in interference-alignment-based cognitive radio networks. IEEE Trans. Veh. Technol.65, 3700–3714 (2016)

6. Zhao, N., Yu, F.R., Leung, V.C.M.: Wireless energy harvesting in interference alignment networks. IEEE Commun. Mag.53, 72–78 (2015)

7. Zhao, N., Yu, F.R., Sun, H.: Adaptive energy-efficient power allocation in green interference-alignment-based wireless networks. IEEE Trans. Veh. Technol. 64, 4268–4281 (2015)

8. Chen, S., Cheng, R.S.: Clustering for interference alignment in multiuser interfer- ence network. IEEE Trans. Veh. Technol.63, 2613–2624 (2014)

9. Zhao, N., Zhang, X., Yu, F.R., Leung, V.C.M.: To align or not to align: topology management in asymmetric interference networks. IEEE Trans. Veh. Technol.

10. Zhang, X., Zhao, N., Yu, F.R., Jin, M., Leung, V.C.M.: Resource allocation in topology management of asymmetric interference networks. IEEE Syst. J.

11. Zhao, N., Yu, F.R., Leung, V.C.M.: Opportunistic communications in interference alignment networks with wireless power transfer. IEEE Wirel. Commun.22, 88–95 (2015)

12. Sanayei, S., Nosratinia, A.: Antenna selection in MIMO systems. IEEE Commun.

Mag.42, 68–73 (2004)

13. Koltz, J.G., Sezgin, A.: Antenna selection criteria for interference alignment. In:

Proceedings of the IEEE PIMRC 2010, pp. 527–531. IEEE Press, Instanbul (2010) 14. Li, X., Zhao, N., Sun, Y., Yu, F.R.: Interference alignment based on antenna selec- tion with imperfect channel state information in cognitive radio networks. IEEE Trans. Veh. Technol.65, 5497–5511 (2016)

15. Zhao, N., Yu, F.R., Sun, H., Nallanathan, A., Yin, H.: A novel interference align- ment scheme based on sequential antenna switching in wireless networks. IEEE Trans. Wirel. Commun.12, 5008–5021 (2013)

16. Rubinstein, R.Y., Kroese, D.P.: The Cross-Entropy Method: A Unified Approach to Combinatorial Optimization, Monte-Carlo Simulation and Machine Learning.

Springer, New York (2004).https://doi.org/10.1007/978-1-4757-4321-0

17. Zhang, Y., Ji, C., Malik, W.Q., O’Brien, D.C., Edwards, D.J.: Receive antenna selection for MIMO systems over correlated fading channels. IEEE Trans. Wirel.

Commun.8, 4393–4399 (2009)

18. Ali, A.O.D., Yetis, C.M., Torlak, M.: Receive antenna selection for MIMO sys- tems over correlated fading channels. In: 2014 IEEE 25th Annual International Symposium on Personal, Indoor, and Mobile Radio Communication (PIMRC), pp.

402–406. IEEE Press, Washington (2014)

Wireless Information and Power Transfer for Multiuser OFDM Systems

Xin Liu1(&), Xiaotong Li2, Zhenyu Na2, and Qiuyi Cao1

1 School of Information and Communication Engineering, Dalian University of Technology, Dalian 116024, China liuxinstar1984@dlut.edu.cn, cccqiu_yi@163.com

2 School of Information Science and Technology, Dalian Maritime University, Dalian 116026, China 565856998@qq.com, nazhenyu@dlmu.edu.cn

Abstract. Most of existing works on simultaneous wireless information and power transfer (SWIPT) for OFDM systems are studied based on power splitting or time splitting, which may lead to the time delay and the decreasing of sub- carrier utilization. In this paper, a multiuser orthogonal frequency division mul- tiplexing (OFDM) system is proposed, which divides the sub-carriers into two parts, one for information decoding and the other one for energy harvesting. We investigate the optimization problem for maximizing the sum rate of users under the constraint of energy harvesting through optimizing the channel allocation and power allocation. By using the iterative algorithm, the optimal solution to the optimization problem can be achieved. The simulation results show that the proposed algorithm converges fast and outperforms the conventional algorithm.

Keywords: SWIPTOFDMSubcarrier allocationPower allocation

1 Introduction

Orthogonal frequency division multiplexing (OFDM) is a viable air interface for providing ubiquitous communication services and high spectral efficiency, due to its ability to combat frequency selective multipath fading and flexibility in resource allocation. However, power-hungry circuitries and the limited energy supply in por- table devices remain the bottlenecks in prolonging the lifetime of networks and guaranteeing quality of service (QoS). As a result, energy-efficient mobile communi- cation has attracted considerable interest from both industry and academia [1–4].

Traditionally, energy has been harvested from natural renewable energy sources such as solar, wind, and geothermal heat, thereby reducing substantially the reliance on the energy supply from conventional energy sources. As a result, simultaneous wireless information and power transfer (SWIPT) is emerged. In [5], the concept of SWIPT is first put forward and the capacity-energy function is defined. Two classical models are put forward in paper [6,7], including time switching (TS) model and power switching (PS) model. In TS model, the receiver switches into energy harvesting mode or

©ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2018 X. Gu et al. (Eds.): MLICOM 2017, Part I, LNICST 226, pp. 524–531, 2018.

https://doi.org/10.1007/978-3-319-73564-1_52

information mode within one transmission time. In PS model, the receiver splits the power into two parts with some ratio, one for information decoding and the other one for energy harvesting. SWIPT is combined with multiple-input-single-output (MISO) in [8], where a transmitter with multi-antenna transmits the same information to several banks of single antenna simultaneously. In [9], the optimization algorithm of power splitting based on down-link OFDMA is proposed by the iterative algorithm. A tradeoff between TS and PS is proposed in [10].

Different from the PS and TS models, we study a sub-carrier allocation-based SWIPT for multiuser OFDM systems without a splitter at the receiver. The sub-carriers of each user are separated into information decoding part and energy harvesting part. We address the problem of maximizing the sum rate of users while keeping enough harvested energy. The non-convex problem is solved by an iterative algorithm.

2 System Model

Consider a wireless OFDM down-link system consisting of one cognitive base station (CBS) andKusers. Each user is only equipped with one antenna. LetKdenote the sets ofkusers forkẳf1;2; Kg. The OFDM bandwidth is assumed to be divided into N(N K) channels equally. The sub-carriers set is denoted withNforN= {1,…,n}.

Each sub-carrier must be allocated to only one user. Parts of sub-carriers are used for energy harvesting, and the others are utilized for information decoding simultaneously.

We suppose that the channel power gain on each sub-carrier is always constant in one transmission period time, which is given at the base station. Lethk;nrepresent the gain of thek-th user on then-th sub-carrier. Then the noise power of each sub-carrier can be corrupted by nk, which is modeled as an additive white Gaussian noise (AWGN) random variable with zero mean and variance r2. The total transmission power is limited to power budget P. Therefore, the power allocated on the n-th sub-carrier is denoted asPn. LetSP represent the sub-carriers used for energy harvesting to power transfer. Accordingly, the other sub-carriers used for information decoding is denoted bySI. Hence,SIK represents the sub-carriers used for information transfer onK-th user.

One sub-carrier cannot be used for energy harvesting and transfer information simul- taneously, so we haveSI\SPẳ ;and SI[SPẳN (Fig.1).

base station User 1

User 2

User 3

User 4 User 5

information transfer energy transfer

Fig. 1. System model.

3 Problem Formulation

Our aim is to maximize the sum rate of OFDM down-link under a restricted condition of the minimum harvested energy for each user. Let Bk represent to the minimum harvested energy ofk-th user. Since one sub-carrier can only be allocated to one user, letan;k be a binary channel allocation index. In other words,an;kẳ1 means that the sub-carriernis only allocated to the userkand an;kẳ0 is determined on other terms.

So it is written as

XK

kẳ1ak;nẳ1;8n2N ð1ị

The sum rate of system can be formulated as XK

kẳ1

X

n2SIak;nlog 1þhk;nPn

r2

ð2ị

where n2SI. With energy harvesting efficiencye, the harvested energy during one transmission block for userkis determined by

X

n2SPehk;nPnþr2

ð3ị For8k2K. Therefore, optimization model of maximum sum rate can be written as follows

an;kmax;SI;Pn

XK

kẳ1

X

n2SIak;nlog 1þhk;nPn

r2

s:t:X

n2NPnP SP[SI ẳN SP\SI ẳ ;

XK

kẳ1ak;nẳ1;8n2N ak;n2f0;1g;8k2K;n2N Pn0;8n2N

ð4ị

4 Optimal Solution

Due to the non-convex problem, it is impossible to obtain the optimal solution directly.

In this section a sub-optimal algorithm is proposed for solving this problem.

Wefirstly optimizeak;nwith givenPnandSIð ị, then optimizeSP Pnwith givenak;n andSIð ị, and optimizeSP SIð ịSP with givenak;n and Pn at last. As mentioned above, Pnand SIð ịSP are determined, soak;n is optimized as follows

maxan;k PK

kẳ1ak;nlog 1ỵhk;nr2Pn

;n2SI s:t:PK

kẳ1ak;nẳ1;8n2N ak;n2f0;1g;8k2K;n2N

ð5ị

The problem above is regarded as allocating the sub-carriernto the assigned user for obtaining the maximum sum rate. In other words, the sub-carrier n (n2SI) is allocated to the user k which can get the maximum hk;nPn, i.e., ak;nẳ1;kẳ argmaxk2K hk;nPn andak;nẳ0;8k6ẳk;k2K.

Secondly,Pn is optimized byak;nand SIð ị. In this proposition, the problem canSP be rewritten as

maxPn P

n2SIklog 1þhkr;n2Pn

s:t:P

n2SPehk;nPnþr2 Bk P

n2NPnP Pn0;8n2N

ð6ị

Note that ak;nẳ1;ak;nẳ0;8k6ẳk;k2K: The converted problem is satisfied with convex model. Thus, the Lagrange dual decomposition is employed for solving this problem. The Lagrange dual function is given as follows:

gðb1;b2ị ẳmaxf gPn L Pð ịn ð7ị whereb1;b2 are the Lagrange multipliers and they are determined by the sub-gradient method. Meanwhile,L Pð ịn is expressed as:

L Pð ị ẳn P

n2SIklog 1þhkr;n2Pn

þb1 P

n2SPehk;nPnþr2 Bk

þ

b2 PP

n2NPn

ð8ị

Then the dual problem can be simplified as follows:

minb1;b2gðb1;b2ị

s:t b1;b20 ð9ị

Because the dual problem is differentiable, it can be solved with the sub-gradient method. The sub-gradient is shown as follow:

Db1ẳX

n2SPehk;nPnỵr2Bk ð10ị

Db2ẳPX

n2NPn ð11ị

For givenb1;b2, the optimal powerPnðn2SIịis obtained by KKT conditions by using mathematical manipulation, as follows

Pnẳ 1 b2r2

hk þ

ð12ị

whereð ị: ỵ denotes max(.,0). Similarly, the allocated powerPn used for information transfering is determined as:

Pnẳ Pmax hk;ne[b2 Pmin hk;neb2

ð13ị

where Pmax and Pmin represent the maximum and minimum power constraints on information decode respectively. According toPnand an;k,SIð ịSP can be obtained by substituting (11) and (12) into (8). Consequently, Lagrange dual function can be rewritten as

L SP ẳXK

kẳ1

X

n2Nan;klog 1þhk;nPn

r2

XK

kẳ1

X

n2SPan;klog 1þhk;nPn

r2

þb1X

n2SPehk;nPnþr2

b1Bkþb2Pb2X

n2NPn

ẳX

n2SP b1ehk;nPnþr2 XK

kẳ1an;klog 1ỵhk;nPn r2

þX

n2N

XK

kẳ1an;klog 1ỵhk;nPn

r2

b2Pn

b1Bkþb2P

ẳX

n2SPFnþX

n2N

XK

kẳ1an;klog 1ỵhk;nPn r2

b2Pn

b1Bkþb2P

ð14ị where

Fnẳb1ehk;nPnỵr2 XK

kẳ1an;klog 1ỵhk;nPn r2

ð15ị

Analyzing the formulate (13), only thefirst item on the right side is about to SP. Thus, the optimalSP can be achieved by maximum the itemFn, as follows

SP ẳarg maxSP

X

n2SPFn ð16ị

SPcan be easily gotten by substituting all thenintoFntofind the ones which are makeFnpositive, then the rest of the setNare belongs toSI. The proposed algorithm to solve the optimal problem is listed as the Algorithm 1.

5 Simulation Results

In this section, the performance of the proposed algorithm based on multi-users SWIPT is demonstrated by simulation results.

We denote all the channels involved are following Rayleigh fading with unit mean.

For simplicity, we assume that the minimum harvested energy limits for all the users are the same, i.e., BkẳB. In addition, we set N = 16, K = 5, r2 = 1, Pmax= P/N, Pmin= 0, andeẳ1.

Figure2shows the convergence behavior of the proposed algorithm. It is seen that the proposed algorithm converges fast, which indicates that the proposed algorithm can be implemented practically. Figure3shows that the comparison between the proposed optimization algorithm and the conventional algorithm. It can be seen that the proposed algorithm performance better compared with conventional algorithm. The conventional allocated N sub-carriers to K users. By contrast, all the sub-carriers are used for information decoding and the consumed energy comes from the system. In the con- ventional algorithm, the water-filling approach is used for power allocation, this will cause some power waste. Moreover, all the sub-carriers are allocated to information decoding which results the energy consumption and less power can be used for information decoding. Figure3also shows that the sum rate of users increases as the sum transmit power P increases. This is because with the same target harvested energy, when the sum transmission power increases there will be more power allocated for information decoding.

Figure4shows that the total transmission power used for information decoding of userk. It can be seen that the user 5 is allocated the most power and user 2 is the least.

That is because in our emulation, the user 5 has the best channel condition which can achieve higher sum rate.

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 15

16 17 18 19 20 21 22 23 24

Number iterations

sum rate of users (bps)

Fig. 2. Convergence behavior of the proposed algorithm.

65 70 75 80 85 90 95 100 105 110

16 18 20 22 24 26 28 30 32

sum transmition power (w)

sum rate of users (bps)

Comparison of performance

The proposed algorithm Convertional algorithm

Fig. 3. The sum transmit power vs sum rate of users

1 2 3 4 5

0 2 4 6 8 10 12 14 16 18 20

user k

Transmission power used for information decoding of user k (W)

Fig. 4. The power used for information decoding of userk

6 Conclusion

In this paper, we propose a joint optimization algorithm for SWIPT-based multi-user OFDM systems. Specifically, the OFDM sub-carriers of each user are divided into two parts, one for information decoding and the other one for energy harvesting. Therefore, we can obtain enough information rate without using time or power splitter at the receiver, on the premise of harvesting enough energy. The simulation results show that the proposed algorithm converges fast and outperforms the conventional algorithm.

Acknowledgments. This work was supported by the National Natural Science Foundations of China under Grant No. 61601221, the Natural Science Foundations of Jiangsu Province under Grant No. BK20140828, the China Postdoctoral Science Foundations under Grant No. 2015M580425 and the Fundamental Research Funds for the Central Universities under Grant No. DUT16RC(3)045.

References

1. Chen, T., Yang, Y., Zhang, H., Kim, H., Horneman, K.: Network energy saving technologies for green wireless access networks. IEEE Wirel. Commun.18, 30–38 (2011)

2. Arnold, O., Richter, F., Fettweis, G., Blume, O.: Power consumption modeling of different base station types in heterogeneous cellular networks. In: Proceedings of Future Network and Mobile Summit, pp. 1–8 (2010)

3. Yang, J., Ulukus, S.: Optimal packet scheduling in an energy harvesting communication system. IEEE Trans. Commun.60, 220–230 (2012)

4. Ng, D.W.K., Lo, E., Schober, R.: Energy-efficient resource allocation in OFDMA systems with large numbers of base station antennas. IEEE Trans. Wirel. Commun.11, 3292–3304 (2012)

5. Varshney, L.R.: Transporting information and energy simultaneously. In: International Symposium on IEEE Information Theory, ISIT 2008, pp. 1612–1616. IEEE (2008) 6. Liu, L., Zhang, R., Ho, C.K.: Wireless information transfer with opportunistic energy

harvesting. IEEE Trans. Wirel. Commun.12(1), 288–300 (2013)

7. Liu, L., Zhang, R., Ho, C.K.: Wireless information and power transfer: a dynamic power splitting approach. IEEE Trans. Wirel. Commun.61(9), 3990–4001 (2013)

8. Ng, D.W.K., Schober, R.: Spectral efficient optimization in OFDM systems with wireless information and power transfer. In: 21st European Signal Processing Conference (EUSIPCO 2013), Marrakech, pp. 1–5 (2013)

9. Zhou, X., Zhang, R., Ho, C.K.: Wireless information and power transfer in multiuser OFDM systems. In: 2013 IEEE Global Communications Conference (GLOBECOM), Atlanta, GA, pp. 4092–4097 (2013)

10. Ng, D.W.K., Lo, E.S., Schober, R.: Wireless information and power transfer: energy efficiency optimization in OFDMA systems. IEEE Trans. Wirel. Commun. 12(12), 6352–

6370 (2013)

Intelligent Ad-Hoc and Sensor Networks

A 100 MHz SRAM Design in 180 nm Process

Zhuangguang Chen and Bei Cao(&)

Electronic Engineering School, Heilongjiang University, Harbin 150080, China caobei@hlju.edu.cn

Abstract. With the development of integrated circuit, SoC systems are more and more used in products. Memory is an important part of SoC, SRAM design is a key research area. In this paper, based on ASIC design methodology, 2 K-bits SRAM is designed. A 6T-SRAM memory cell is designed and simu- lated with circuit level to improve reliability. The memory cell is used to con- struct the storage array, which are the word line 32 bits and the bit line 8 bits.

Then, the SRAM peripheral circuit is designed and simulated by using SMIC 0.18lm process, including the data input/output buffer circuit, clock circuit, address decoding circuit, data read/write circuit and sense amplifier. The structure, function and performance of latch type sense amplifier are analyzed emphatically. The simulation results demonstrate that the function of SRAM is verified correctly. The clock frequency of the SRAM can reach 100 MHz.

Keywords: Static random access memoryMemory arrayPeripheral circuits

1 Introduction

With the rapid development of digitalization process, the information industry is going into the era of large data. SoC chip has become the key of information and data processing. The data storage in SoC is an important part. In order to solve the storage problems of SoC data storage, improve the working performance of SoC processor, and reduce the speed gap between processor and external memory, memory hierarchical technology is adopted in this paper. Static random access memory (SRAM) with high speed and low capacity is used as the key technology to solve the above problems. In this paper, with the analysis of the internal working mechanism about SRAM, the design methodology based on ASIC is adopted, in order to achieve 2568 bytes.

A variety of design implementation and improvement schemes are proposed for SRAM. A novel power gated 9T SRAM cell is proposed, which uses read decoupling access buffers and power gated transistors to perform reliable read and write operations [1]. A bit line equivalent scheme is proposed to eliminate the leakage dependence of the data pattern. Thus, the read bit line sensing and its stability to the process, voltage and temperature variations are improved [2]. A 6T SRAM operating down to near threshold regime is presented [3]. A dual-port spin-orbit torque magnetic RAM for on-chip caching applications with reduced power consumption is proposed to reduce the impact of write delay on performance [4].

©ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2018 X. Gu et al. (Eds.): MLICOM 2017, Part I, LNICST 226, pp. 535–544, 2018.

https://doi.org/10.1007/978-3-319-73564-1_53

Một phần của tài liệu Machine learning and intelligent communications part i 2017 (Trang 530 - 575)

Tải bản đầy đủ (PDF)

(722 trang)