Intel's STrategic Offensive Research & Mitigations (STORM) department, which the company set up back in 2017 when it learned of side-channel attack vulnerabilities in its CPUs, have penned a paper detailing a proposed solution to the problem. Intel's offensive security research team counts with around 60 workers who focus on proactive security testing and in-depth investigations. Of that group, STORM is a subset of around 12 individuals who specifically work on prototyping exploits to show their practical impact. The solution proposed by this group is essentially a new memory-based hardware fix, going by the name of SAPM (Speculative-Access Protected Memory). The new solution would implement a resistant hardware fix in the CPU's memory that essentially includes blocks for known speculative-access hacks, such as the ones that hit Intel CPUs hard such as Meltdown, Foreshadow, MDS, SpectreRSB and Spoiler.

For now, the proposed solution is only at a "theory and possible implementation options" level. It will take a long time for it to find its way inside working Intel CPUs - if it ever does, really, since for now, it's just a speculative solution. A multitude of tests have to be done in order for its implementation to be approved and finally etched into good old silicon. Intel's STORM says that the SAPM approach would carry a performance hit; however, the group also calculates it to be "potentially lesser" than the current impact of all released software mitigations. Since the solution doesn't address every discovered side-channel attack specifically, but addresses the type of back-end operations that concern these attacks, the team is confident this solution would harden Intel CPUs against (most of) both known and not-yet-known speculative execution hacks.

As the industry's leading provider of SSD controllers and storage solutions, Goke Microelectronics was invited to the 2019 Flash Memory Summit to demonstrate an ultra-low latency NVMe SSD based on Toshiba Memory's XL-FLASH memory. One year ago, Toshiba Memory announced XL-FLASH at the 2018 Flash Memory Summit, promising to use ultra-low latency 3D SLC flash to reduce read latency to 5μs, which is equivalent to 1/10th of read latency of 3D TLC NAND.

Goke 2311-series drives are based on the 2311 SSD controller and are paired with Toshiba Memory's XL-FLASH memory. The prototype of 2311-series drives have implemented an overall 4K random read latency under 20μs and the final drives will offer a 4K random read latency in less than 15μs. Goke 2311-series drives support up to 4 TB capacity with a maximum write bandwidth of 1 GB/s and read bandwidth of 3 GB/s through a PCIe Gen 3 x4 interface. They will also support SM2/3/4 and SHA-256/AES-256 with built-in security engines.

CORSAIR, a world leader in PC gaming peripherals and enthusiast components, today announced the availability of a new kit of its award-winning CORSAIR VENGEANCE LPX DDR4 Memory, setting a new record for commercially available high-frequency DRAM with clock speeds up to an incredible 4,866 MHz. This blazingly fast memory is now available in a 2x 8 GB kit, reaching its full potential in systems with new 3rd Gen AMD Ryzen Desktop Processors and X570 motherboards.

VENGEANCE LPX has long pushed the limits of performance memory, and continues that tradition today thanks to the partnership between CORSAIR and AMD. While fully compatible with most modern platforms, the new modules have been specifically designed and fully tested to achieve their record-breaking maximum frequency of 4,866 MHz on the Ryzen 3000 platform in ASUS ROG Crosshair VIII Formula, MSI MEG X570 GODLIKE, and MSI PRESTIGE X570 CREATION motherboards using their included automated overclocking utilities.

G.SKILL, the world's leading manufacturer of extreme performance memory and gaming peripherals, is very thrilled to announce a new world record for the fastest memory frequency at DDR4-6016.8 MHz, being the first ever to break through the DDR4-6000 barrier. This milestone is set by the Taiwanese professional overclocker, TopPC, using G.SKILL DDR4 Trident Z Royal memory on the latest MSI MPG Z390I GAMING EDGE AC motherboard and Intel Core i9-9900K processor.

Just 3 short months after the world record DDR4-5886MHz set by Toppc at the G.SKILL Computex booth in June, Toppc has once again set a new world record for memory frequency speed and is the first to push DDR4 memory to an astonishing 6 GHz. The validation links for this major milestone can be found in the following CPU-Z validation link and HWBot website screenshot below (validation).

Samsung's B die has been widely known as a good, high performance variant of DRAM memory, loved by overclockers because of its ability to get to a high frequency with relatively low timings. However, B die has been discontinued and now Samsung started offering its replacement in form of the newly developed A die manufactured in 1z nm (1z class) lithography process. Despite the lack of technical details surrounding the new die type, Hardwareluxx has received a tip from its reader about new RAM offering that incorporates A die memory.

The M378A4G43AB2-CVF, as it is called in the listing, is a 32 GB, single dimm DDR4 RAM with operating speed of 2933 MHz and CL21-21-21 timings. This particular offer isn't something to be excited about as the frequency is good, but the timings are quite high for that speed. Given that we don't know where the A die is targeted at, we can speculate that its current aim is at mid-tier systems, where the mediocre performance is okay and the system isn't suffering (performance wise) because of it. Nonetheless this find is quite interesting as it gives first hints at what can we expect in therms of future A die DRAM offerings. Remember, it took some time for B die as well to get to the level of performance we have today, so it is entirely possible that A die will improve and try to aim for greater performance level than it currently has.

G.SKILL International Enterprise Co., Ltd., the world's leading manufacturer of extreme performance memory and gaming peripherals, is excited to announce high-frequency, high-capacity memory kit specifications to the G.SKILL Trident Z Royal series at DDR4-4300 MHz CL19-19-19-39 and DDR4-4000 MHz CL16-18-18-38, both in kit capacity of 64 GB (8x 8 GB) and 32 GB (4x8GB). Once again, these extraordinary G.SKILL Trident Z Royal memory kits are built with high performance Samsung B-die components.

As an excellent choice for tasks that require both high-capacity and extreme speed memory, the high-performance DDR4-4300 MHz CL19-19-19-39 64 GB (8x 8 GB) is shown in the stress test screenshot below, validated on the latest ASUS ROG Rampage VI Extreme Encore motherboard and the Intel Core i9-9820X processor. This is the perfect memory solution for building an all-powerful workstation for various content creation workloads, including 3D rendering and multimedia editing.

"Control" by Remedy is the season's hottest AAA release, not just because it's an above-average story-driven action RPG, but also because it's an eye candy-shop. With the ability to use NVIDIA RTX real-time raytracing across a multitude of features, the game is particularly heavy on graphics hardware. Tweaktown tested the game's stability at extremely high display resolutions, including 8K, and found that the game can use up to 18.5 GB of video memory, when running in DirectX 12 with RTX enabled. There's only one client-segment graphics card capable of that much memory, the $2,499 NVIDIA TITAN RTX, which ships with 24 GB of GDDR6 memory. Its nearest client-segment neighbor is the AMD Radeon VII, but it only packs 16 GB of HBM2.

When a game needs more video memory than your graphics card has, Windows has an elaborate memory management system that sheds some of that memory onto your system's main memory, and the swap file progressively (at reduced performance, of course). Video memory usage drops like a rock between 8K and 4K UHD (which is 1/4th the pixels as 8K). With all RTX features enabled and other settings maxed out, "Control" only uses 8.1 GB of video memory. What this also means is that video cards with just 8 GB of memory are beginning fall short of what it takes to game at 4K. The $699 GeForce RTX 2080 Super only has 8 GB. The RTX 2080 Ti, with its 11 GB of memory has plenty of headroom and muscle. Find other interesting observations in the source link below.

Yangtze Memory Technologies (YMTC), a Chinese state-backed semiconductor company founded in 2016 as part of the Chinese Government's tech-independence push, has commenced mass-production of 64-layer 3D NAND flash memory chips, at a rate of 100,000 to 150,000 wafers per month leading into 2020. The 64-layer 3D NAND chips are based on YMTC's "in-house" Xtracking architecture. The company is already developing a 128-layer 3D NAND flash chip, and is skipping 96-layer along the way.

YMTC's capacity will be augmented by a new fab being built by its parent company, Tsinghua Unigroup. Tsinghua is a state-owned company which holds a controlling 51 percent stake in YMTC, and is a beneficiary of China's National Semiconductor Industry Investment Fund. When it goes online in 2021-22, the new Tsinghua fab, located in Chengdu, will augment YMTC's capacity by an additional 100,000 12-inch wafers per month. Its existing fab in Nanjing will also receive a capacity expansion.

AMD's next-generation "Renoir" APU, which succeeds the company's 12 nm "Picasso," will be the company's truly next-generation chip to feature an integrated graphics solution. It's unclear as of now, if the chip will be based on a monolithic die, or if it will be a multi-chip module of a 7 nm "Zen 2" chiplet paired with an enlarged I/O controller die that has the iGPU. We're getting confirmation on two key specs - one, that the iGPU will be based on the older "Vega" graphics architecture, albeit with an updated display engine to support the latest display standards; and two, that the processor's memory controller will support the latest LPDDR4X memory standard, at speeds of up to 4266 MHz DDR. In comparison, Intel's "Ice Lake-U" chip supports LPDDX4X up to 3733 MHz.

Code-lines pointing toward "Vega" graphics with an updated display controller mention the new DCN 2.1, found in AMD's new "Navi 10" GPU. This controller supports resolutions of up to 8K, DSC 1.2a, and new resolutions of 4K up to 240 Hz and 8K 60 Hz over a single cable, along with 30 bits per pixel color. The multimedia engine is also suitably updated to VCN 2.1 standard, and provides hardware-accelerated decoding for some of the newer video formats, such as VP9 and H.265 at up to 90 fps at 4K, and 8K up to 24 fps, and H.264 up to 150 fps at 4K. There's no word on when "Renoir" comes out, but a 2020 International CES unveil is likely.

Toshiba Memory Holdings Corporation, which will rebrand as Kioxia Holdings Corporation on October 1, 2019, announced today that it has signed a definitive agreement with LITE-ON Technology Corporation to acquire its Solid State Drive (SSD) business. The purchase price is 165 million US dollars; the transaction is expected to close by the first half of 2020 and is subject to customary closing adjustments and regulatory approval.

LITE-ON is a Taiwan-based supplier of optoelectronics, storage, semiconductors and other devices. Toshiba Memory and LITE-ON share a commitment to quality, innovation and manufacturing excellence. With cultural synergies and LITE-ON's proven experience in the SSD field for personal computers and data centers, Toshiba Memory sees this acquisition as a way to significantly strengthen its SSD business.

Toshiba at the Flash Memory Summit announced they've managed to develop a 5-Bit-per-Cell memory solution Based on its BiCS 4 flash memory technologies, the feat was achieved using a modified module of Quad-Level Cell (QLC) memory. This shows the technology is not only feasible, but has room for improvement, since an adapted QLC technology will always be inferior to a natively-developed, Penta-Level Cell (PLC) solution.

To achieve this ability to store one extra bit of information per cell (compared to QLC), a new level of voltage refinement is required: the cell has to be able to change its state according to one of 32 voltage states, which, in turn, have to be read out correctly by the flash memory controller. This reduces the cell's performance and endurance (as does any increase in the number of bits per cell), and will require a number of solutions to mitigate and compensate for this reduced performance. However, density has become an increasing concern from manufacturers, hence the continued development of deeper, more variable voltage states that allow for even more information to be stored in the same silicon area. Higher density means cheaper solutions, but density increased in such a way has known trade-offs that have been much talked about ever since the transition from Single-Level Cell (SLC) up to the (nowadays ubiquitous) QLC.

HyperX, the gaming division of Kingston Technology, Inc., today announced the release of FURY DDR4 RGB and a new look to the FURY DDR4 lineup. The new memory kits both offer automatic Plug N Play overclocking functionality. For DDR4 users that need fast, high-density memory, HyperX has added models up to 64 GB at 3466 MHz. The new FURY DDR4 and FURY DDR4 RGB are cost-effective, high-performance upgrades for Intel and AMD's latest platforms that feature Plug N Play, which enables automatic memory overclocking at standard DDR4 1.2 V settings. The new FURY models also feature Intel XMP-ready profiles optimized for Intel's latest platform.

The HyperX FURY DDR4 RGB comes equipped with an LED light bar with fluid RGB lighting effects and utilizes HyperX Infrared Sync, which allows the modules to remain synchronized. The memory is compatible with lighting control software from a range of motherboard vendors, including ASUS Aura Sync, Gigabyte RGB Fusion, and MSI Mystic Light Sync. Additional information about software and motherboard compatibility is available online at respective partner websites. FURY DDR4 RGB memory is also compatible with HyperX NGenuity software.

Samsung Electronics, Ltd., has taken its leadership position in the memory market a step further today by announcing support of the Samsung PM1733 PCIe Gen4 Solid State Drive (SSD) and high density RDIMM and LRDIMM dynamic random access memory (DRAM) for the AMD EPYC 7002 Generation Processors. AMD launched the 2nd Gen AMD EPYC processor in San Francisco yesterday.

"AMD has listened to the needs of its customers in developing the 2nd Gen AMD EPYC processors and has worked closely with us to integrate the best of our cutting-edge memory and storage products," said Jinman Han, senior VP of Memory Product Planning, Samsung Electronics. "With these new datacenter processors, AMD is providing customers with a processor that enables a new standard for the modern datacenter."

Samsung Electronics, the world leader in advanced memory technology, today announced that it has begun mass producing 250-gigabyte (GB) SATA solid state drive (SSD) that integrates the company's sixth-generation (1xx-layer) 256-gigabit (Gb) three-bit V-NAND for global PC OEMs. By launching a new generation of V-NAND in just 13 months, Samsung has reduced the mass production cycle by four months while securing the industry's highest performance, power efficiency and manufacturing productivity.

"By bringing cutting-edge 3D memory technology to volume production, we are able to introduce timely memory lineups that significantly raise the bar for speed and power efficiency," said Kye Hyun Kyung, executive vice president of Solution Product & Development at Samsung Electronics. "With faster development cycles for next-generation V-NAND products, we plan to rapidly expand the markets for our high-speed, high-capacity 512 Gb V-NAND-based solutions."

Toshiba Memory America, Inc. (TMA), the U.S.-based subsidiary of Toshiba Memory Corporation, today announced the launch of a new Storage Class Memory (SCM) solution: XL-FLASH. Based on the company's innovative BiCS FLASH 3D flash memory technology with 1-bit-per-cell SLC, XL-FLASH brings low latency and high performance to data center and enterprise storage. Sample shipments will start in September, with mass production expected to begin in 2020.

Classified as SCM (or persistent memory), with the ability to retain its contents like NAND flash memory, XL-FLASH bridges the performance gap that exists between DRAM and NAND. While volatile memory solutions such as DRAM provide the access speed needed by demanding applications, that performance comes at a high cost. As the cost-per-bit and scalability of DRAM levels off, this new SCM (or persistent memory) layer in the memory hierarchy addresses that issue with a high density, cost effective, non-volatile NAND flash memory solution. Poised for growth, industry analyst firm IDC estimates the SCM market is expected to reach in excess of $3B in 2022.

Marvell today released the industry's lowest power PCIe Gen4 NVMe solid-state drive (SSD) controller portfolio. Marvell's newest SSD controllers are designed to meet the need for lower power and higher performance in next-generation data centers and edge devices as artificial intelligence (AI) and 5G gain momentum. This breakthrough technology delivers unparalleled performance in an ultra-compact footprint, leveraging the company's complex system-on-chip (SoC) design expertise and groundbreaking storage IP to help data center, notebook, tablet, gaming and edge computing platform architects advance their solutions for the highly distributed data era.

"Marvell's latest family of storage controllers has been architected to optimally address edge computing and data center pain points of power-performance and capacity-performance," said Nigel Alvares, vice president of marketing for the Flash Business Unit at Marvell Semiconductor, Inc. "With today's launch, we're once again demonstrating Marvell's leadership in storage, delivering the industry's first 4-Channel PCIe Gen4 NVMe SSD controllers with the industry's lowest power consumption that will help revolutionize SSD solutions for the data economy."

G.SKILL International Enterprise Co., Ltd., the world's leading manufacturer of extreme performance memory and gaming peripherals, is releasing a highly optimized, extremely low latency Trident Z Neo series DDR4-3800MHz CL14 RGB memory kit in 16 GB (8 GB x2) and 32 GB (8 GB x4) capacities for the AMD Ryzen 3000 series CPU and X570 chipset platform. Built with the powerful Samsung B-die component, this is the perfect DDR4 memory kit for those looking to push the limits of memory bandwidth on your new AMD Ryzen 3000 platform.

At this point, it's well-known that memory performance with the new AMD Ryzen 3000 processor series is best when Infinity Fabric being tied to the memory clock at a 1:1 ratio. The G.SKILL R&D team is dedicated to push the performance boundaries even further and developed a high-frequency, low-latency memory kit at DDR4-3800 CL14-16-16-36 in capacity configurations of 8GBx2 and 8GBx4, reaching a superb memory bandwidth performance under the optimal 1:1 ratio.

SK Hynix Inc. announced today that it has developed and starts mass-producing the world's first 128-Layer 1 Tb (Terabit) TLC (Triple-Level Cell) 4D NAND Flash, only eight months after the Company announced the 96-Layer 4D NAND Flash last year.

The 128-Layer 1 Tb NAND chip offers the industry's highest vertical stacking with more than 360 billion NAND cells, each of which stores 3 bits, per one chip. To achieve this, SK Hynix applied innovative technologies, such as "ultra-homogeneous vertical etching technology," "high-reliability multi-layer thin-film cell formation technology," and ultra-fast low-power circuit design, to its own 4D NAND technology.

Prices of DRAM shot up by 20 percent as Japan put in place export curbs that restrict high-technology exports to South Korea, and as Toshiba recovers from a power blackout that temporarily halted production. This could impact prices of end-user products such as PC memory modules, or consumer electronics, such as smartphones, in the coming weeks, as inventories either dry up, or are marked-up at various stages of the supply-chain. The memory industry is inter-dependent between fabrication and packaging units spread across South Korea, Japan, and Taiwan.

Memory and flash industry observer DRAMeXchange reported that spot-pricing of 8-gigabit DDR4 DRAM chips, which is used as a benchmark for DRAM pricing as a whole, closed at USD $3.74 at the end of trading on Friday (19/07). It's up 14.6 percent week-over-week, and 23 percent up pricing as on 5th July. An industry observer who spoke with KBS World notes that the recent hikes are not directly infuenced by the trade-spat between Japan and Korea, but rather a power blackout experienced at a Toshiba DRAM manufacturing facility last month. The observer noted that if the trade-spat affects production at Samsung Electronics or SK Hynix, DRAM prices could "skyrocket."

CORSAIR, a world leader in PC gaming peripherals and enthusiast components, today announced the addition of 32 GB modules to its range of VENGEANCE LPX high-performance DDR4 memory, allowing PC builders to equip their systems with more DDR4 memory than ever before. VENGEANCE LPX has long been a premiere choice for custom PC builders looking for high frequencies and ambitious overclocks, and that tradition continues with the launch of 32 GB modules - the first time that such a capacity of premium DRAM has been made widely available to consumers in a standard size DDR4 module.

The new modules feature the same craftsmanship and quality that CORSAIR customers expect from the VENGEANCE LPX name. Thoroughly tested for wide compatibility with most current DDR4 motherboards, designed for high-performance overclocking with a pure aluminium heatspreader, and available in multiple colors to match your system's look, VENGEANCE LPX 32 GB DDR4 modules set the standard for enthusiast memory. Launching in frequencies of 2,400 MHz and 2,666 MHz in kits of 1x, 2x,4x and 8x modules, or 3,000 MHz in kits of 1x and 2x modules, you'll be sure to find a configuration to fit your custom PC and take its memory capacity up to 128 GB on mainstream 4-DIMM slot, and up to 256 GB on high-end desktop 8-DIMM slot motherboards.

GeIL, Golden Emperor International Ltd. - one of the world's leading PC components & peripheral manufacturers released new additions to their EVO X and EVO X ROG-certified RGB Gaming Memory today. The new EVO X II and EVO X II ROG-Certified inherit their previous generation, and provide an advanced modder-friendly "cableless RGB illumination design" for offering a fascinating RGB lighting effect without hassles of power cable management. Manufactured with outstanding components, the EVO X II dedicated to unlocking hidden overclocking ability and comprehensive compatibility across the latest 3rd Gen AMD Ryzen and Intel platforms. GeIL EVO X II ROG-certified RGB Gaming Memory perfectly supports ASUS AURA lighting control app, when paired with most ASUS ROG motherboards, an automatic performance boost will be enabled for a better overclocking experience.

The EVO X DDR4 series is one of the first RGB DDR4 memory modules in the industry, and it brought the series tremendous praise from gamers and PC enthusiasts around the world. The new EVO X II and EVO X II ROG-certified equipped with "cableless RGB illumination design" offering RGB lighting without power cables or the need for advanced cable management. The EVO X II is available in a new stealth black and frost white design and features high-quality aluminium for better heat dissipation and extra stability.

NVIDIA's upcoming GeForce RTX 2080 Super graphics card doesn't just max out the 12 nm "TU104" silicon and add higher GPU clock-speeds, but also features the highest-clocked GDDR6 memory solution on the market, to make the most of the 256-bit wide memory bus of the silicon. NVIDIA deployed 15.5 Gbps GDDR6 memory, which is 10.7 percent faster than the 14 Gbps memory used on the original RTX 2080 and other RTX 20-series graphics cards. The memory real-clock is set at 1937 MHz compared to 1750 MHz on the original RTX 2080. At this memory frequency, the RTX 2080 Super enjoys a memory bandwidth just a touch short of 500 GB/s, at 496 GB/s.

Besides memory, the RTX 2080 Super maxes out the "TU104" silicon by enabling all 3,072 CUDA cores physically present, as opposed to just 2,944 of them being enabled on the original RTX 2080. The card is also endowed with 192 TMUs, 64 ROPs, 384 Tensor cores, and 48 RT cores. The GPU frequencies are set at 1650 MHz with 1815 MHz GPU Boost, compared to 1515/1710 MHz of the original RTX 2080. NVIDIA is launching the RTX 2080 Super at an MSRP of USD $699, with availability slated for July 23. The company's add-in card (AIC) partners are allowed to design custom-design cards that come with improved cooling solutions and higher clocks.

G.SKILL International Enterprise Co., Ltd., the world's leading manufacturer of extreme performance memory and gaming peripherals, is thrilled to announce the launch of Trident Z Neo DDR4 memory series for the latest AMD Ryzen 3000 series CPUs on AMD X570 platform. Featuring optimized specifications for the new AMD platform, a new sleek dual-tone heatspreader design, and fully customizable RGB lighting, the Trident Z Neo memory series is the ideal DDR4 DRAM for your next AMD gaming system or workstation.

Optimized Performance for AMD Ryzen 3000: Computer systems with AMD Ryzen processors are known for its performance scaling with memory speed, and Trident Z Neo is engineered to achieve optimal performance with the latest AMD Ryzen 3000 series processors on AMD X570 motherboards. Under the latest AMD Ryzen 3000 series platform, DDR4 memory frequency support has increased by leaps and bounds, allowing the X570 chipset platform to run an unprecedented memory speed record of DDR4-5774MHz- the fastest memory speed ever achieved on an AMD platform under extreme liquid nitrogen cooling - as seen in the following CPU-Z validation screenshot with the MSI MEG X570 GODLIKE motherboard.

According to Nikkei, the newly established trade limitations between Japan and South Korea, could end up compromising global memory supply by simply restricting chemicals export. As the report says, Japan has limited its export of three vital chemicals (like orthophosphoric, hydrobromic and citric acid) used in semiconductor manufacturing, to South Korea.

Unlike before, a company that exports a chemical, now has to ask for a permission from the Japanese government, so it could supply the semiconductor foundries in South Korea. The end result of such move could be severely damaged global memory supply, as over 70% of DRAM and over 50% of NAND memory is manufactured in South Korea. Government processing of applications for exporting chemicals is estimated to take about three months, while the memory makers usually only hold around one of two months of extra supply for manufacturing. SK Hynix, a third biggest memory manufacturer by revenue, said that if it doesn't get enough stock of materials, it would have to halt production. Samsung is asserting the situation for now, without any elaboration on that. These events could lead to increased memory price and overall less supply.

Scaling and manufacturing of ever shrinking semiconductor devices is becoming more challenging as smaller nodes are introduced. As we have approached 7 nanometers, economies of scale are becoming more influential than scales of manufacturing. For example, the development of the 7 nm node development cost more than 3 billion USD, while smaller nodes are expected to see that price cross the 5 billion USD mark. So given that we are approaching the limit where we can't squeeze more transistors in two-dimensional space without huge economical impact, we have to utilize another dimension in order to keep performance improvements coming.

AMD has filed a patent for cooling a 3D stacked memory with thermo-electric coolers - TECs, also known as Peltier devices. Being that TECs are made out of P-type and N-type semiconductors, they can easily be integrated into existing silicon manufacturing methods and controlled like a regular device. The process AMD has patented basically describes how to insert the TEC between memory and logic devices, where it draws heat from either logic or memory with each side being able to dissipate the heat. That effect is possible due to nature of TEC, where the direction of heat flow is changed inverting the voltage.