Optical RAM Report

Optical Random Access Memory

Department of Electronics and Communication, SJCE

1. Introduction

The exponential growth of internet traffic has been the major driving force for the increasing demand for transmission bandwidth. To increase the efficiency of the network and to allow high data bit rates it is desirable that switching and routing can be carried out in the physical layer, avoiding optical-to-electrical and electrical-to-optical converters. All-optical devices provide data format transparency, and may provide lower power consumption and higher-speed processing, compared to their electronic counterparts. Recent developments in optical signal processing and in photonic switching have made it possible to reach bit rates in the order of gigabits per second per wavelength and terabits per second per fibre. In this context, all-optical flip-flops (AOFF) can be used to perform many optical signal processing functions in future optical packet switching networks.

2. Need for Optical RAM

The evolving trend of chip multiprocessor (CMPs) utilization for extensive use of parallelism along with multi core architectures have accelerated the demand of high-bandwidth and low-latency interconnects in High-Performance Computing Systems (HPCS). Recent improvements in interconnects have equipped data centers with energy efficient inter- and intra-chip communication, resulting to powerful machines with up to 40 Gb/s line-rate capabilities and Tb/s throughput [6] . However, given that HPCS are already entering the Peta-Flops regime , a new set of challenges are expected to strain the data center networking infrastructure as the increase of the processor throughput is exceeding the improvement rate of memory speeds. From a design perspective, a balanced system requires equal improvements in all functional subsystems besides data transmission gateways, including memory blocks.

However, electronic RAM has proven incapable of keeping pace with current processing speeds. Limited memory bandwidth and long access times can degrade overall system performance, forming a major bottleneck in system’s performance that is commonly referred to as the “Memory Wall” [2]. Memory needs of digital processing and computing has been formed so far around the idiosyncrasy of electronic RAM. Between types of electronic memories, static RAMs have been the preferred choice for performance-sensitive applications, implementing cache memories on the processor chips of HPCS.

To overcome the long foreseen “Memory Wall”, research has shifted focus on developing all optical RAM alternatives based on ultra-fast bistable latching and memory elements. Figure 1 shows that there has been intensive research on optical flip-flops in the last



Figure 1: IEEE Xplore annual citations with index terms referring to optical RAM.

few years. This fact can be explained mainly due to technology advances in monolithic and hybrid integration, and the need for a high-speed and low-power memory.

3. Types of Optical Memory 3.1 Optical Read-Only Memory

In Optical ROM, data is stored on the disc as a series of microscopic indentations.

A laser is shone onto the reflective surface of the disc to read the pattern of pits and lands ("pits", with the gaps between them referred to as "lands"). Because the depth of the pits is approximately one-quarter to one-sixth of the wavelength of the laser light used to read the disc, the reflected beam's phase is shifted in relation to the incoming beam, causing destructive interference and reducing the reflected beam's intensity. This pattern of changing intensity of the reflected beam is converted into binary data. Examples are Compact Disc, Digital Video Disc, Blu-ray Disc. 3.2 Optical Random Access Memory

Optical RAM allows stored data to be accessed directly in any random order. There are two types of RAM. 3.2.1 Dynamic random-access memory (DRAM)

It is a type of random-access memory that stores each bit of data in a separate capacitor within an integrated circuit. The capacitor can be either charged or discharged; these two states are taken to represent the two values of a bit, conventionally called 0 and 1. Since capacitors leak charge, the information eventually fades unless the



capacitor charge is refreshed periodically. Because of this refresh requirement, it is a dynamic memory as opposed to SRAM and other static memory. An optical DRAM can store information in its optical format for a long term and can be demonstrated with a refreshable WDM loop memory in place of capacitors. 3.2.2 Static random-access memory (SRAM)

It is a type of semiconductor memory that uses bistable latching circuitry to store each bit. The term static differentiates it from dynamic RAM (DRAM) which must be periodically refreshed. SRAM exhibits data remanence, but it is still volatile in the conventional sense that data is eventually lost when the memory is not powered. Optical S-RAM is implemented using Semiconductor Optical Amplifier (SOA) and Mach-Zehnder Interferometer (MZI). 4. Comparison of SRAM and DRAM

Compared to DRAM, SRAM does not have a capacitor to store the data, hence SRAM works without refreshing.

In SRAM the data is lost when the memory is not electrically powered. SRAM is faster and more reliable than the more common DRAM. While DRAM supports access times (access time is the time required to read or write

data to/from memory) of about 60 nanoseconds, SRAM can give access times as low as 10 nanoseconds.

Cycle time of SRAM is much shorter than that of DRAM because it does not need to pause between accesses.

SRAM is much more expensive to produce than DRAM. The operation of SRAM and DRAM are the same in both optical and electrical domains.

Only difference lies in the implementation of the fundamental block. 5. Basic S-RAM Block

A typical 2-dimensional arrangement of 4X4 static RAM bank is shown in Figure 2, where 16 separate single RAM cells are independently controlled and each row can store a 4-bit word. Shared among the RAM cells of a single row, the “Word” signal grants simultaneous access for Read or Write operation according to the logical value of the corresponding signal.

The most basic element in electronic RAM bank structures for static memory design has been the 6-transistor (6T) RAM cell, the layout of which is shown in Figure 3. It consists of two



pass gates for access control and two cross-coupled inverters with two possible states. Although it has been widely used in electronic cache memories, its speed performance imposes the major limit to the overall system processing speed.

Figure 2: 4X4 static RAM bank architecture Figure 3: 6 Transistor electronic RAM Cell

Optical memory has to overcome innate limitations enforced by the neutral charge of light particles that impedes their storage, making it impossible to mimic the respective electronic memory modules that rely on the negative electron charge. Optical bit-level storage without random access capabilities has been recently demonstrated by means of integrated optical memory elements relying on coupled semiconductor lasers.

The proposed optical RAM cell consists of two Access Gates (AG) and a hybridly integrated all-optical flip-flop (AOFF) as shown in figure 4, experimentally demonstrating successful operation at 5Gb/s along with a performance analysis for reaching 40Gb/s Read/Write speeds

Figure 4: Basic All-Optical RAM cell [1]



6. Implementation of all-optical SRAM

The access transistors in electronic SRAM are replaced by access gates which is nothing but a Semiconductor Optical Amplifier. Cross coupled inverters are replaced by an All-Optical Flip-Flop. The way in which AOFF is implemented depends on the application. There are many ways of implementing AOFF. Some are discussed below. 6.1 Coupled Ring Lasers and Arrayed Waveguide AOFF

Here, the AOFF is implemented using two coupled ring lasers and an arrayed waveguide grating (AWG). AWG acts as the frequency selective element of each cavity as shown in figure 5(a). Each ring laser has its own gain element (SOA) and the light propagates bidirectionally in the ring. Its operation principle is shown in figure 5(b) and is based on the gain quenching concept.

-

Figure 5: a) Mask layout for two-state multi wavelength laser. b) Operation of device showing laser [3]

In state 1, light from laser A flows in counterclockwise direction and saturates the gain

element of laser B, suppressing laser B from lasing. In this situation, the light from the dominant laser (laser A) is sent to an output by the AWG. Thus, the AWG ensure isolation between the device input and output.

In the same way, in state 2, only laser B is lasing, suppressing laser A from lasing. Thus, only one of the lasers can be lasing at a time, due of the fact that the dominant laser suppresses



the other laser through gain saturation. Lasing in the master can be turned off by injecting external optical pulses, changing in this way the state of the system. The cavity length of this AOFF scheme is about 4.5 mm, which cause speed limitation. 6.2 Coupled SOA-MZI AOFF

AOFF can also be implemented with Mach-Zehnder interferometers (MZIs) as the coupled nonlinear optical elements and nonlinear polarization switches [4]. Figure 6 shows all-optical flip-flops demonstrated using two coupled Mach-Zehnder interferometers with semiconductor optical amplifier (SOA-MZI) and two SOA fibre ring lasers, respectively. This technology is also based on the gain quenching effect, in which the signal output from the dominant laser suppresses the other laser, through gain saturation of the SOA.

Figure 6: Schematic diagram of optical flip-flop based on coupled SOA-MZI

The SOAs operate as Cross-Gain Modulation (XGM) switches controlling access to the flip-flop configuration, whereas the optical flip-flop serves as the 1-bit memory element. Error-free Read and Write functionality with true random access properties at 5 Gb/s can be achieved in the optical domain. 6.3 Micro ring LASER based AOFF

In a micro-ring laser typically there exists two lasing modes, based on the direction that laser light propagates- clockwise (CLKW) or counterclockwise (CCLKW) and depending on the bias current, different operating regimes can occur. Also, ideally, the CLKW mode is not coupled to the CCLKW mode and vice-versa.



Figure 7: Optical flip-flop based on single micro-ring laser

The AOFF configuration is depicted in Figure 7. It has racetrack cavity geometry and it is produced in an InGaAs/InGaA1As/InP Multi Quantum Well material. The AOFF exhibits bistability between the counter-propagation cavity modes that can be switched by external optical pulses. It has four inputs/outputs ports that are all-active, and can be use to inject or extract optical signals. 7. Comparison between various AOFF technologies Table 1: Comparison between various AOFF technologies

Parameter\Technology Ring lasers and AWG SOA-MZI Single Micro Ring Laser

Switching Times < 1 ns < 200 ps < 190 ps

Switching Energies < 10 pJ < 1 pJ 4 fJ

Extinction Ratio 35 dB SM > 10 dB DM > 20 dB SM

Integration Type Monolithic Hybrid Monolithic

*Results shown above are for Asynchronous S-R Flip-Flops.



8. Applications of All-Optical Flip-Flops

AOFFs are analogous to electronic flip-flops in optical domain, which implies that all the circuits using electronic flip-flops can be implemented in optical domain using AOFFs

An all-optical flip-flop is an element that can be used to perform a variety of optical signal processing functions that requires memory, such as a binary counter, optical shift register and optical random access memory, among others. 8.1 All-Optical Signal Processing

All-optical signal processing is expected to overcome the limitations of electrical signal processing and to lead optical communications to the next level. Therefore, it is essential to develop all-optical sequential devices which allow an effective improvement of the all-optical processing capabilities of future networks.

An all-optical flip-flop is an element that can be used to perform a variety of optical signal processing functions that requires memory, such as a binary counter, optical shift register and optical random access memory, among others. 8.1.1 All-Optical Binary Counter

Figure 8: Logic circuit for all-optical binary counter



An all-optical binary counter can be used for header recognizing and payload processing in optical packet-switching networks, besides working as a finite-state machine in optical computing. The experimental setup [7] is shown in Figure 8. The scheme is composed by two identical stages realized each one by an optical S-R latch and an optical AND logic gate based on FWM in SOA.

Each stage has 3 ports: the input clock, the output (Qi) and the carry signal. The latches output Q1 and Q2, represent, respectively, the less significant bit (LSB) and most significant bit (MSB) of the counter output. In order to carry out two-bit counting operations, the two stages are cascaded by using the "Carry 1" signal coming from stage 1, as input for the second stage. Extension to N-bit counting can be performed by cascading N stages. 8.1.2 All-Optical Shift Register

A shift register is a device capable of retaining information in the form of a binary number, i.e., the register can store a binary word. In addition of being used as temporary memory, an optical shift register also have the ability to make successive shifts to the left or to the right. An optical shift register scheme [8] is demonstrated in figure 9, which consists of an optical converter in combination with two cascaded AOFFs.

Figure 9: Optical shift register scheme

The intensity modulated input data is transformed by the optical converter into wavelength encoded data and, subsequently, injected into the port ‘In1’ of AOFF1, which is cascaded with AOFF2. Each AOFF is based on two ring lasers, with a single active element and a feedback arm [14]. The two cascaded AOFFs are controlled by optical clock pulses that are required to clear the states of the optical flip-flops. The output from AOFF1 then sets the new



state of AOFF2. After this operation, the delayed clock pulse is injected into the port ‘In2’ of AOFF1 and subsequently clears its state. The signal encoded in wavelength that outputs from the optical converter then sets the new state of AOFF1. Thus a compete ‘shift’ function has been realized. 8.2 All-Optical RAM

An all-optical RAM eliminates the need for optical-electrical-optical (O/E/O) conversion, which Is the main cause of delays in the case of opto-electronic RAMs, where a large portion of the time and energy is wasted due to repeated opto-Electronic conversions. Basic AO-RAMs support error free Read/Write Speeds of 5 Gbps. More Complex AO-RAM systems making use of WDM have been shown to support speeds of up to 160 Gbps (compared to 2 Gbps supported by comparable e-RAMs)

8.3 All-Optical Packet Routing An optical packet routing/switching scheme can forward optical packets to the appropriate destination port, based on the address information that is encoded by the attached labels. The information required for optical routing is carried together with the optical signal (e.g. label) and since it can be extracted and added from the optical signal without optical-electrical-optical (O/E/O) conversion, lower power consumptions can be expected 8.3.1 All-Optical Packet Switching

The optical flip-flop based on a single SOA-MZI, with a coupler inside the feedback loop, can be used as an optical packet-forwarding unit, depending on the optical set/reset information.

The setup [9] of the proposed packet switching scheme is shown in Figure 10 and data flow is forwarded through the flip-flop, without compromise the latching operation. The operation principle of this packet switching scheme is very simple: the packet comes out if the flip-flop is ON, and is blocked otherwise.

Accordingly, if a packet is delimited with valid header and trailer, the address recognition module will deliver a Set pulse at the beginning of the packet and a Reset pulse at the end of packet. The first pulse will turn the flip-flop ON and allow the packet to propagate to its destination. The second pulse will turn it OFF so that further packets with invalid addresses are blocked.



Figure 10: Packet forwarding gate based on flip-flop

8.3.2 All-Optical Content Resolution

The all-optical packet switching schemes, presented in 8.3.2, can forward one packet at any given moment. However, when multiple packets arrive for the same output port, at the same time, contention occurs. All-optical packet contention can be solved in the wavelength, space and time domain. In the wavelength domain, wavelength conversion is needed to prevent the collision of the packets. In the space domain, space deflection provides separate routes, which avoids the contention of the packets. Finally, in the time domain, when packets arrive at the same time, they are routed to different optical delay lines (ODLs) and the collision is avoided.

In figure 11, an optical circuit is demonstrated, relying on the hybrid integrated optical flip-flop that performs on-the-fly contention resolution between 40 Gb/s packets of the same wavelength.

Figure 11: Contention resolution concept [10]



9. Advantages of Optical Flip-Flops over Electronic Flip-Flops Increased circuit simplicity Increased Speed Reduced number of active elements and associated power consumption RAM bank implementations with smarter column/row encoders/decoder while enabling for

re-configurability in optical cache mapping. 10. Disadvantages of Optical Flip-Flops Difficulty in fabrication since related technologies are still under research As of now, commercially not viable. Full potential cannot be exploited as of now. 11. Conclusion

Despite the proven high-speed potential of optical signal processing circuitry, photonic processing devices still experience several difficulties in convincing about their functional potential, one main reason for this being the absence of a reliable optical Random Access Memory (RAM).

As network traffic continues to increase each year, unless Moore’s law reaches to a limit, the power consumption in CMOS VLSI chips will soon consume hundreds of Watts. Optical communication networks, with photonic integration capability, may provide higher speeds, with reduced environmental impacts. Applications nowadays range from all-optical shift registers to threshold functions and packet forwarding, being packet duration holding the most common and immediate application. Integration and silicon photonics can bring a new age of development for these devices in terms of cost and performance integration in more complex functions, opening in this way a route for a next generation of applications and functionalities.



REFERENCES [1] George T. Kanellos, Dimitrios Fitsios, Theonitsa Alexoudi, Christos Vagionas, Amalia Miliou and Nikos Pleros, “ Bringing WDM into optical static RAM architectures”,Lightwave technology journal, March. 2013 [2] S. McKee, “Reflections on the Memory Wall,” in Proceedings of the 1st Conf. on Comp. frontiers, Ischia, Italy, Apr. 2004. [3] Hill M., Vries T., Dorren H. J. S., Leijtens X., Zantvoort J. H. C., Besten J., Smallbrugge B., Oei Y., Binsma H., Khoe G., Smit M.: ‘Integrated two-state AWG-based multiwavelength laser’. IEEE Photon. Technol. Letters, 2005. [4] M.T. Hill, H. de Waardt, G.D. Khoe, and H.J.S. Dorren, "Fast optical flip-flop by use of Mach-Zender Interferometers", Microwave Opt. Technol. Lett., vol. 31, no. 6, pp. 411-415, Dec. 2001 [5] Trita A., Mezosi G., Bragheri F., Jin Y., Furst S., Elsasser W., Cristiani I., Sorel M., Giuliani G.: ‘Dynamic operation of all-optical flip-flop based on a monolithic semiconductor ring laser’. ECO’08 Proc., 2008, paper We2C3, Brussels [6] “Photonic RAM: The key for facilitating an Optical Router development”, NTT Photonic Laboratories report. [7] Wang J., Meloni G., Berrettini G., Poti L., Bogoni A..: ‘All-optical binary counter based on Semiconductor Optical Amplifiers’.Op. Letters, 2009, 34, (22), Pp. 3517–3519 [8] Zhang S., Li Z., Liu Y., Khoe G., Dorren H. J. S.: ‘Optical Flip-Flop Shift Register Based On An Optical Flip-Flop Memory With A Single Active Element’. Osa Optics Express, 2005, 13, Pp.9708-9713. [9] Brahmi H., Bougioukos M., Menif M., Maziotis A., Stamatiadis C., Kouloumentas C., Apostolopoulos D., Avramopoulos H., Erasme D.: ‘‘Experimental Demonstration Of An All-Optical Packet Forwarding Gate Based On A Single Soa-Mzi At 40 Gb/S’’. Proc. Ofc/Nfoec 2011, 2011, Paper Omk5, Los Angeles, California [10] Stamatiadis C., Bougioukos M., Maziotis A., Bakopoulos P., Stampoulidis L., Avramopoulos H.: ‘‘All-Optical Contention Resolution using a single optical flip-flop and two stage all-optical wavelength conversion’’. OFC/NFOEC’10, 2010, paper OThN5, San Diego, USA

Optical RAM Report

Documents

Transcript of Optical RAM Report